summaryrefslogtreecommitdiffstats
path: root/runtime/queue.c
blob: 9d7a9058137375415b28b81726a3486688b73a4d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
/* queue.c
 *
 * This file implements the queue object and its several queueing methods.
 * 
 * File begun on 2008-01-03 by RGerhards
 *
 * There is some in-depth documentation available in doc/dev_queue.html
 * (and in the web doc set on http://www.rsyslog.com/doc). Be sure to read it
 * if you are getting aquainted to the object.
 *
 * Copyright 2008 Rainer Gerhards and Adiscon GmbH.
 *
 * This file is part of the rsyslog runtime library.
 *
 * The rsyslog runtime library is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * The rsyslog runtime library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public License
 * along with the rsyslog runtime library.  If not, see <http://www.gnu.org/licenses/>.
 *
 * A copy of the GPL can be found in the file "COPYING" in this distribution.
 * A copy of the LGPL can be found in the file "COPYING.LESSER" in this distribution.
 */
#include "config.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <signal.h>
#include <pthread.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/stat.h>	 /* required for HP UX */
#include <time.h>
#include <errno.h>

#include "rsyslog.h"
#include "queue.h"
#include "stringbuf.h"
#include "srUtils.h"
#include "obj.h"
#include "wtp.h"
#include "wti.h"
#include "msg.h"
#include "atomic.h"

#ifdef OS_SOLARIS
#	include <sched.h>
#	define pthread_yield() sched_yield()
#endif

/* static data */
DEFobjStaticHelpers
DEFobjCurrIf(glbl)
DEFobjCurrIf(strm)

/* forward-definitions */
static rsRetVal qqueueChkPersist(qqueue_t *pThis);
static rsRetVal qqueueSetEnqOnly(qqueue_t *pThis, int bEnqOnly, int bLockMutex);
static rsRetVal qqueueRateLimiter(qqueue_t *pThis);
static int qqueueChkStopWrkrDA(qqueue_t *pThis);
static int qqueueIsIdleDA(qqueue_t *pThis);
static rsRetVal qqueueConsumerDA(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave);
static rsRetVal qqueueConsumerCancelCleanup(void *arg1, void *arg2);
static rsRetVal qqueueUngetObj(qqueue_t *pThis, obj_t *pUsr, int bLockMutex);

/* some constants for queuePersist () */
#define QUEUE_CHECKPOINT	1
#define QUEUE_NO_CHECKPOINT	0

/* methods */


/* get the overall queue size, which includes ungotten objects. Must only be called
 * while mutex is locked!
 * rgerhards, 2008-01-29
 */
static inline int
qqueueGetOverallQueueSize(qqueue_t *pThis)
{
#if 0 /* leave a bit in for debugging -- rgerhards, 2008-01-30 */
BEGINfunc
dbgoprint((obj_t*) pThis, "queue size: %d (regular %d, ungotten %d)\n",
	   pThis->iQueueSize + pThis->iUngottenObjs, pThis->iQueueSize, pThis->iUngottenObjs);
ENDfunc
#endif
	return pThis->iQueueSize + pThis->iUngottenObjs;
}


/* This function drains the queue in cases where this needs to be done. The most probable
 * reason is a HUP which needs to discard data (because the queue is configured to be lossy).
 * During a shutdown, this is typically not needed, as the OS frees up ressources and does
 * this much quicker than when we clean up ourselvs. -- rgerhards, 2008-10-21
 * This function returns void, as it makes no sense to communicate an error back, even if
 * it happens.
 */
static inline void queueDrain(qqueue_t *pThis)
{
	void *pUsr;
	
	ASSERT(pThis != NULL);

	/* iQueueSize is not decremented by qDel(), so we need to do it ourselves */
	while(ATOMIC_DEC_AND_FETCH(pThis->iQueueSize) > 0) {
		pThis->qDel(pThis, &pUsr);
		if(pUsr != NULL) {
			objDestruct(pUsr);
		}
	}
}


/* --------------- code for disk-assisted (DA) queue modes -------------------- */


/* returns the number of workers that should be advised at
 * this point in time. The mutex must be locked when
 * ths function is called. -- rgerhards, 2008-01-25
 */
static inline rsRetVal qqueueAdviseMaxWorkers(qqueue_t *pThis)
{
	DEFiRet;
	int iMaxWorkers;

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(!pThis->bEnqOnly) {
		if(pThis->bRunsDA) {
			/* if we have not yet reached the high water mark, there is no need to start a
			 * worker. -- rgerhards, 2008-01-26
			 */
			if(qqueueGetOverallQueueSize(pThis) >= pThis->iHighWtrMrk || pThis->bQueueStarted == 0) {
				wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* disk queues have always one worker */
			}
		} else {
			if(pThis->qType == QUEUETYPE_DISK || pThis->iMinMsgsPerWrkr == 0) {
				iMaxWorkers = 1;
			} else {
				iMaxWorkers = qqueueGetOverallQueueSize(pThis) / pThis->iMinMsgsPerWrkr + 1;
			}
			wtpAdviseMaxWorkers(pThis->pWtpReg, iMaxWorkers); /* disk queues have always one worker */
		}
	}

	RETiRet;
}


/* wait until we have a fully initialized DA queue. Sometimes, we need to
 * sync with it, as we expect it for some function.
 * rgerhards, 2008-02-27
 */
static rsRetVal
qqueueWaitDAModeInitialized(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ASSERT(pThis->bRunsDA);

	while(pThis->bRunsDA != 2) {
		d_pthread_cond_wait(&pThis->condDAReady, pThis->mut);
	}

	RETiRet;
}


/* Destruct DA queue. This is the last part of DA-to-normal-mode
 * transistion. This is called asynchronously and some time quite a
 * while after the actual transistion. The key point is that we need to
 * do it at some later time, because we need to destruct the DA queue. That,
 * however, can not be done in a thread that has been signalled 
 * This is to be called when we revert back to our own queue.
 * This function must be called with the queue mutex locked (the wti
 * class ensures this).
 * rgerhards, 2008-01-15
 */
static rsRetVal
qqueueTurnOffDAMode(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ASSERT(pThis->bRunsDA);

	/* at this point, we need a fully initialized DA queue. So if it isn't, we finally need
	 * to wait for its startup... -- rgerhards, 2008-01-25
	 */
	qqueueWaitDAModeInitialized(pThis);

	/* if we need to pull any data that we still need from the (child) disk queue,
	 * now would be the time to do so. At present, we do not need this, but I'd like to
	 * keep that comment if future need arises.
	 */

	/* we need to check if the DA queue is empty because the DA worker may simply have
	 * terminated do to no new messages arriving. That does not, however, mean that the
	 * DA queue is empty. If there is still data in that queue, we do nothing and leave
	 * that for a later incarnation of this function (it will be called multiple times
	 * during the lifetime of DA-mode, depending on how often the DA worker receives an
	 * inactivity timeout. -- rgerhards, 2008-01-25
	 */
	if(pThis->pqDA->iQueueSize == 0) {
		pThis->bRunsDA = 0; /* tell the world we are back in non-DA mode */
		/* we destruct the queue object, which will also shutdown the queue worker. As the queue is empty,
		 * this will be quick.
		 */
		qqueueDestruct(&pThis->pqDA); /* and now we are ready to destruct the DA queue */
		dbgoprint((obj_t*) pThis, "disk-assistance has been turned off, disk queue was empty (iRet %d)\n",
			  iRet);
		/* now we need to check if the regular queue has some messages. This may be the case
		 * when it is waiting that the high water mark is reached again. If so, we need to start up
		 * a regular worker. -- rgerhards, 2008-01-26
		 */
		if(qqueueGetOverallQueueSize(pThis) > 0) {
			qqueueAdviseMaxWorkers(pThis);
		}
	}

	RETiRet;
}


/* check if we run in disk-assisted mode and record that
 * setting for easy (and quick!) access in the future. This
 * function must only be called from constructors and only
 * from those that support disk-assisted modes (aka memory-
 * based queue drivers).
 * rgerhards, 2008-01-14
 */
static rsRetVal
qqueueChkIsDA(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	if(pThis->pszFilePrefix != NULL) {
		pThis->bIsDA = 1;
		dbgoprint((obj_t*) pThis, "is disk-assisted, disk will be used on demand\n");
	} else {
		dbgoprint((obj_t*) pThis, "is NOT disk-assisted\n");
	}

	RETiRet;
}


/* Start disk-assisted queue mode. All internal settings are changed. This is supposed
 * to be called from the DA worker, which must have been started before. The most important
 * chore of this function is to create the DA queue object. If that function fails,
 * the DA worker should return with an appropriate state, which in turn should lead to
 * a re-set to non-DA mode in the Enq process. The queue mutex must be locked when this
 * function is called, else a number of races will happen.
 * Please note that this function may be called *while* we in DA mode. This is due to the
 * fact that the DA worker calls it and the DA worker may be suspended (and restarted) due
 * to inactivity timeouts.
 * rgerhards, 2008-01-15
 */
static rsRetVal
qqueueStartDA(qqueue_t *pThis)
{
	DEFiRet;
	uchar pszDAQName[128];

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pThis->bRunsDA == 2) /* check if already in (fully initialized) DA mode... */
		FINALIZE;       /* ... then we are already done! */

	/* create message queue */
	CHKiRet(qqueueConstruct(&pThis->pqDA, QUEUETYPE_DISK , 1, 0, pThis->pConsumer));

	/* give it a name */
	snprintf((char*) pszDAQName, sizeof(pszDAQName)/sizeof(uchar), "%s[DA]", obj.GetName((obj_t*) pThis));
	obj.SetName((obj_t*) pThis->pqDA, pszDAQName);

	/* as the created queue is the same object class, we take the
	 * liberty to access its properties directly.
	 */
	pThis->pqDA->pqParent = pThis;

	CHKiRet(qqueueSetpUsr(pThis->pqDA, pThis->pUsr));
	CHKiRet(qqueueSetsizeOnDiskMax(pThis->pqDA, pThis->sizeOnDiskMax));
	CHKiRet(qqueueSetiDeqSlowdown(pThis->pqDA, pThis->iDeqSlowdown));
	CHKiRet(qqueueSetMaxFileSize(pThis->pqDA, pThis->iMaxFileSize));
	CHKiRet(qqueueSetFilePrefix(pThis->pqDA, pThis->pszFilePrefix, pThis->lenFilePrefix));
	CHKiRet(qqueueSetiPersistUpdCnt(pThis->pqDA, pThis->iPersistUpdCnt));
	CHKiRet(qqueueSetbSyncQueueFiles(pThis->pqDA, pThis->bSyncQueueFiles));
	CHKiRet(qqueueSettoActShutdown(pThis->pqDA, pThis->toActShutdown));
	CHKiRet(qqueueSettoEnq(pThis->pqDA, pThis->toEnq));
	CHKiRet(qqueueSetEnqOnly(pThis->pqDA, pThis->bDAEnqOnly, MUTEX_ALREADY_LOCKED));
	CHKiRet(qqueueSetiDeqtWinFromHr(pThis->pqDA, pThis->iDeqtWinFromHr));
	CHKiRet(qqueueSetiDeqtWinToHr(pThis->pqDA, pThis->iDeqtWinToHr));
	CHKiRet(qqueueSetiHighWtrMrk(pThis->pqDA, 0));
	CHKiRet(qqueueSetiDiscardMrk(pThis->pqDA, 0));
	if(pThis->toQShutdown == 0) {
		CHKiRet(qqueueSettoQShutdown(pThis->pqDA, 0)); /* if the user really wants... */
	} else {
		/* we use the shortest possible shutdown (0 is endless!) because when we run on disk AND
		 * have an obviously large backlog, we can't finish it in any case. So there is no point
		 * in holding shutdown longer than necessary. -- rgerhards, 2008-01-15
		 */
		CHKiRet(qqueueSettoQShutdown(pThis->pqDA, 1));
	}

	iRet = qqueueStart(pThis->pqDA);
	/* file not found is expected, that means it is no previous QIF available */
	if(iRet != RS_RET_OK && iRet != RS_RET_FILE_NOT_FOUND)
		FINALIZE; /* something is wrong */

	/* as we are right now starting DA mode because we are so busy, it is
	 * extremely unlikely that any regular worker is sleeping on empty queue. HOWEVER,
	 * we want to be on the safe side, and so we awake anyone that is waiting
	 * on one. So even if the scheduler plays badly with us, things should be
	 * quite well. -- rgerhards, 2008-01-15
	 */
	wtpWakeupWrkr(pThis->pWtpReg); /* awake all workers, but not ourselves ;) */

	pThis->bRunsDA = 2;	/* we are now in DA mode, but not fully initialized */
	pThis->bChildIsDone = 0;/* set to 1 when child's worker detect queue is finished */
	pthread_cond_broadcast(&pThis->condDAReady); /* signal we are now initialized and ready to go ;) */

	dbgoprint((obj_t*) pThis, "is now running in disk assisted mode, disk queue 0x%lx\n",
		  qqueueGetID(pThis->pqDA));

finalize_it:
	if(iRet != RS_RET_OK) {
		if(pThis->pqDA != NULL) {
			qqueueDestruct(&pThis->pqDA);
		}
		dbgoprint((obj_t*) pThis, "error %d creating disk queue - giving up.\n", iRet);
		pThis->bIsDA = 0;
	}

	RETiRet;
}


/* initiate DA mode
 * param bEnqOnly tells if the disk queue is to be run in enqueue-only mode. This may
 * be needed during shutdown of memory queues which need to be persisted to disk.
 * If this function fails (should not happen), DA mode is not turned on.
 * rgerhards, 2008-01-16
 */
static inline rsRetVal
qqueueInitDA(qqueue_t *pThis, int bEnqOnly, int bLockMutex)
{
	DEFiRet;
	DEFVARS_mutexProtection;
	uchar pszBuf[64];
	size_t lenBuf;

	BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, bLockMutex);
	/* check if we already have a DA worker pool. If not, initiate one. Please note that the
	 * pool is created on first need but never again destructed (until the queue is). This
	 * is intentional. We assume that when we need it once, we may also need it on another
	 * occasion. Ressources used are quite minimal when no worker is running.
	 * rgerhards, 2008-01-24
	 */
	if(pThis->pWtpDA == NULL) {
		lenBuf = snprintf((char*)pszBuf, sizeof(pszBuf), "%s:DA", obj.GetName((obj_t*) pThis));
		CHKiRet(wtpConstruct		(&pThis->pWtpDA));
		CHKiRet(wtpSetDbgHdr		(pThis->pWtpDA, pszBuf, lenBuf));
		CHKiRet(wtpSetpfChkStopWrkr	(pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int)) qqueueChkStopWrkrDA));
		CHKiRet(wtpSetpfIsIdle		(pThis->pWtpDA, (rsRetVal (*)(void *pUsr, int)) qqueueIsIdleDA));
		CHKiRet(wtpSetpfDoWork		(pThis->pWtpDA, (rsRetVal (*)(void *pUsr, void *pWti, int)) qqueueConsumerDA));
		CHKiRet(wtpSetpfOnWorkerCancel	(pThis->pWtpDA, (rsRetVal (*)(void *pUsr, void*pWti)) qqueueConsumerCancelCleanup));
		CHKiRet(wtpSetpfOnWorkerStartup	(pThis->pWtpDA, (rsRetVal (*)(void *pUsr)) qqueueStartDA));
		CHKiRet(wtpSetpfOnWorkerShutdown(pThis->pWtpDA, (rsRetVal (*)(void *pUsr)) qqueueTurnOffDAMode));
		CHKiRet(wtpSetpmutUsr		(pThis->pWtpDA, pThis->mut));
		CHKiRet(wtpSetpcondBusy		(pThis->pWtpDA, &pThis->notEmpty));
		CHKiRet(wtpSetiNumWorkerThreads	(pThis->pWtpDA, 1));
		CHKiRet(wtpSettoWrkShutdown	(pThis->pWtpDA, pThis->toWrkShutdown));
		CHKiRet(wtpSetpUsr		(pThis->pWtpDA, pThis));
		CHKiRet(wtpConstructFinalize	(pThis->pWtpDA));
	}
	/* if we reach this point, we have a "good" DA worker pool */

	/* indicate we now run in DA mode - this is reset by the DA worker if it fails */
	pThis->bRunsDA = 1;
	pThis->bDAEnqOnly = bEnqOnly;

	/* now we must now adivse the wtp that we need one worker. If none is yet active,
	 * that will also start one up. If we forgot that step, everything would be stalled
	 * until the next enqueue request.
	 */
	wtpAdviseMaxWorkers(pThis->pWtpDA, 1); /* DA queues alsways have just one worker max */

finalize_it:
	END_MTX_PROTECTED_OPERATIONS(pThis->mut);
	RETiRet;
}


/* check if we need to start disk assisted mode and send some signals to
 * keep it running if we are already in it. It also checks if DA mode is
 * partially initialized, in which case it waits for initialization to
 * complete.
 * rgerhards, 2008-01-14
 */
static inline rsRetVal
qqueueChkStrtDA(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);

	/* if we do not hit the high water mark, we have nothing to do */
	if(qqueueGetOverallQueueSize(pThis) != pThis->iHighWtrMrk)
		ABORT_FINALIZE(RS_RET_OK);

	if(pThis->bRunsDA) {
		/* then we need to signal that we are at the high water mark again. If that happens
		 * on our way down the queue, that doesn't matter, because then nobody is waiting
		 * on the condition variable.
		 * (Remember that a DA queue stops draining the queue once it has reached the low
		 * water mark and restarts it when the high water mark is reached again - this is
		 * what this code here is responsible for. Please note that all workers may have been
		 * terminated due to the inactivity timeout, thus we need to advise the pool that
		 * we need at least one).
		 */
		dbgoprint((obj_t*) pThis, "%d entries - passed high water mark in DA mode, send notify\n",
			  qqueueGetOverallQueueSize(pThis));
		qqueueAdviseMaxWorkers(pThis);
	} else {
		/* this is the case when we are currently not running in DA mode. So it is time
		 * to turn it back on.
		 */
		dbgoprint((obj_t*) pThis, "%d entries - passed high water mark for disk-assisted mode, initiating...\n",
			  qqueueGetOverallQueueSize(pThis));
		qqueueInitDA(pThis, QUEUE_MODE_ENQDEQ, MUTEX_ALREADY_LOCKED); /* initiate DA mode */
	}

finalize_it:
	RETiRet;
}


/* --------------- end code for disk-assisted queue modes -------------------- */


/* Now, we define type-specific handlers. The provide a generic functionality,
 * but for this specific type of queue. The mapping to these handlers happens during
 * queue construction. Later on, handlers are called by pointers present in the
 * queue instance object.
 */

/* -------------------- fixed array -------------------- */
static rsRetVal qConstructFixedArray(qqueue_t *pThis)
{
	DEFiRet;

	ASSERT(pThis != NULL);

	if(pThis->iMaxQueueSize == 0)
		ABORT_FINALIZE(RS_RET_QSIZE_ZERO);

	if((pThis->tVars.farray.pBuf = malloc(sizeof(void *) * pThis->iMaxQueueSize)) == NULL) {
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);
	}

	pThis->tVars.farray.head = 0;
	pThis->tVars.farray.tail = 0;

	qqueueChkIsDA(pThis);

finalize_it:
	RETiRet;
}


static rsRetVal qDestructFixedArray(qqueue_t *pThis)
{
	DEFiRet;
	
	ASSERT(pThis != NULL);

	queueDrain(pThis); /* discard any remaining queue entries */

	if(pThis->tVars.farray.pBuf != NULL)
		free(pThis->tVars.farray.pBuf);

	RETiRet;
}


static rsRetVal qAddFixedArray(qqueue_t *pThis, void* in)
{
	DEFiRet;

	ASSERT(pThis != NULL);
	pThis->tVars.farray.pBuf[pThis->tVars.farray.tail] = in;
	pThis->tVars.farray.tail++;
	if (pThis->tVars.farray.tail == pThis->iMaxQueueSize)
		pThis->tVars.farray.tail = 0;

	RETiRet;
}

static rsRetVal qDelFixedArray(qqueue_t *pThis, void **out)
{
	DEFiRet;

	ASSERT(pThis != NULL);
	*out = (void*) pThis->tVars.farray.pBuf[pThis->tVars.farray.head];

	pThis->tVars.farray.head++;
	if (pThis->tVars.farray.head == pThis->iMaxQueueSize)
		pThis->tVars.farray.head = 0;

	RETiRet;
}


/* -------------------- linked list  -------------------- */

/* first some generic functions which are also used for the unget linked list */

static inline rsRetVal qqueueAddLinkedList(qLinkedList_t **ppRoot, qLinkedList_t **ppLast, void* pUsr)
{
	DEFiRet;
	qLinkedList_t *pEntry;

	ASSERT(ppRoot != NULL);
	ASSERT(ppLast != NULL);

	if((pEntry = (qLinkedList_t*) malloc(sizeof(qLinkedList_t))) == NULL) {
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);
	}

	pEntry->pNext = NULL;
	pEntry->pUsr = pUsr;

	if(*ppRoot == NULL) {
		*ppRoot = *ppLast = pEntry;
	} else {
		(*ppLast)->pNext = pEntry;
		*ppLast = pEntry;
	}

finalize_it:
	RETiRet;
}

static inline rsRetVal qqueueDelLinkedList(qLinkedList_t **ppRoot, qLinkedList_t **ppLast, obj_t **ppUsr)
{
	DEFiRet;
	qLinkedList_t *pEntry;

	ASSERT(ppRoot != NULL);
	ASSERT(ppLast != NULL);
	ASSERT(ppUsr != NULL);
	ASSERT(*ppRoot != NULL);
	
	pEntry = *ppRoot;
	*ppUsr = pEntry->pUsr;

	if(*ppRoot == *ppLast) {
		*ppRoot = NULL;
		*ppLast = NULL;
	} else {
		*ppRoot = pEntry->pNext;
	}
	free(pEntry);

	RETiRet;
}

/* end generic functions which are also used for the unget linked list */


static rsRetVal qConstructLinkedList(qqueue_t *pThis)
{
	DEFiRet;

	ASSERT(pThis != NULL);

	pThis->tVars.linklist.pRoot = 0;
	pThis->tVars.linklist.pLast = 0;

	qqueueChkIsDA(pThis);

	RETiRet;
}


static rsRetVal qDestructLinkedList(qqueue_t __attribute__((unused)) *pThis)
{
	DEFiRet;

	queueDrain(pThis); /* discard any remaining queue entries */

	/* with the linked list type, there is nothing left to do here. The
	 * reason is that there are no dynamic elements for the list itself.
	 */

	RETiRet;
}

static rsRetVal qAddLinkedList(qqueue_t *pThis, void* pUsr)
{
	DEFiRet;

	iRet = qqueueAddLinkedList(&pThis->tVars.linklist.pRoot, &pThis->tVars.linklist.pLast, pUsr);
#if 0
	qLinkedList_t *pEntry;

	ASSERT(pThis != NULL);
	if((pEntry = (qLinkedList_t*) malloc(sizeof(qLinkedList_t))) == NULL) {
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);
	}

	pEntry->pNext = NULL;
	pEntry->pUsr = pUsr;

	if(pThis->tVars.linklist.pRoot == NULL) {
		pThis->tVars.linklist.pRoot = pThis->tVars.linklist.pLast = pEntry;
	} else {
		pThis->tVars.linklist.pLast->pNext = pEntry;
		pThis->tVars.linklist.pLast = pEntry;
	}

finalize_it:
#endif
	RETiRet;
}

static rsRetVal qDelLinkedList(qqueue_t *pThis, obj_t **ppUsr)
{
	DEFiRet;
	iRet = qqueueDelLinkedList(&pThis->tVars.linklist.pRoot, &pThis->tVars.linklist.pLast, ppUsr);
#if 0
	qLinkedList_t *pEntry;

	ASSERT(pThis != NULL);
	ASSERT(pThis->tVars.linklist.pRoot != NULL);
	
	pEntry = pThis->tVars.linklist.pRoot;
	*ppUsr = pEntry->pUsr;

	if(pThis->tVars.linklist.pRoot == pThis->tVars.linklist.pLast) {
		pThis->tVars.linklist.pRoot = NULL;
		pThis->tVars.linklist.pLast = NULL;
	} else {
		pThis->tVars.linklist.pRoot = pEntry->pNext;
	}
	free(pEntry);

#endif
	RETiRet;
}


/* -------------------- disk  -------------------- */


static rsRetVal
qqueueLoadPersStrmInfoFixup(strm_t *pStrm, qqueue_t __attribute__((unused)) *pThis)
{
	DEFiRet;
	ISOBJ_TYPE_assert(pStrm, strm);
	ISOBJ_TYPE_assert(pThis, qqueue);
	CHKiRet(strm.SetDir(pStrm, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir())));
finalize_it:
	RETiRet;
}


/* This method checks if we have a QIF file for the current queue (no matter of
 * queue mode). Returns RS_RET_OK if we have a QIF file or an error status otherwise.
 * rgerhards, 2008-01-15
 */
static rsRetVal 
qqueueHaveQIF(qqueue_t *pThis)
{
	DEFiRet;
	uchar pszQIFNam[MAXFNAME];
	size_t lenQIFNam;
	struct stat stat_buf;

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pThis->pszFilePrefix == NULL)
		ABORT_FINALIZE(RS_RET_NO_FILEPREFIX);

	/* Construct file name */
	lenQIFNam = snprintf((char*)pszQIFNam, sizeof(pszQIFNam) / sizeof(uchar), "%s/%s.qi",
			     (char*) glbl.GetWorkDir(), (char*)pThis->pszFilePrefix);

	/* check if the file exists */
	if(stat((char*) pszQIFNam, &stat_buf) == -1) {
		if(errno == ENOENT) {
			dbgoprint((obj_t*) pThis, "no .qi file found\n");
			ABORT_FINALIZE(RS_RET_FILE_NOT_FOUND);
		} else {
			dbgoprint((obj_t*) pThis, "error %d trying to access .qi file\n", errno);
			ABORT_FINALIZE(RS_RET_IO_ERROR);
		}
	}
	/* If we reach this point, we have a .qi file */

finalize_it:
	RETiRet;
}


/* The method loads the persistent queue information.
 * rgerhards, 2008-01-11
 */
static rsRetVal 
qqueueTryLoadPersistedInfo(qqueue_t *pThis)
{
	DEFiRet;
	strm_t *psQIF = NULL;
	uchar pszQIFNam[MAXFNAME];
	size_t lenQIFNam;
	struct stat stat_buf;
	int iUngottenObjs;
	obj_t *pUsr;

	ISOBJ_TYPE_assert(pThis, qqueue);

	/* Construct file name */
	lenQIFNam = snprintf((char*)pszQIFNam, sizeof(pszQIFNam) / sizeof(uchar), "%s/%s.qi",
			     (char*) glbl.GetWorkDir(), (char*)pThis->pszFilePrefix);

	/* check if the file exists */
	if(stat((char*) pszQIFNam, &stat_buf) == -1) {
		if(errno == ENOENT) {
			dbgoprint((obj_t*) pThis, "clean startup, no .qi file found\n");
			ABORT_FINALIZE(RS_RET_FILE_NOT_FOUND);
		} else {
			dbgoprint((obj_t*) pThis, "error %d trying to access .qi file\n", errno);
			ABORT_FINALIZE(RS_RET_IO_ERROR);
		}
	}

	/* If we reach this point, we have a .qi file */

	CHKiRet(strm.Construct(&psQIF));
	CHKiRet(strm.SettOperationsMode(psQIF, STREAMMODE_READ));
	CHKiRet(strm.SetsType(psQIF, STREAMTYPE_FILE_SINGLE));
	CHKiRet(strm.SetFName(psQIF, pszQIFNam, lenQIFNam));
	CHKiRet(strm.ConstructFinalize(psQIF));

	/* first, we try to read the property bag for ourselfs */
	CHKiRet(obj.DeserializePropBag((obj_t*) pThis, psQIF));
	
	/* then the ungotten object queue */
	iUngottenObjs = pThis->iUngottenObjs;
	pThis->iUngottenObjs = 0; /* will be incremented when we add objects! */

	while(iUngottenObjs > 0) {
		/* fill the queue from disk */
		CHKiRet(obj.Deserialize((void*) &pUsr, (uchar*)"msg", psQIF, NULL, NULL));
		qqueueUngetObj(pThis, pUsr, MUTEX_ALREADY_LOCKED);
		--iUngottenObjs; /* one less */
	}

	/* and now the stream objects (some order as when persisted!) */
	CHKiRet(obj.Deserialize(&pThis->tVars.disk.pWrite, (uchar*) "strm", psQIF,
			       (rsRetVal(*)(obj_t*,void*))qqueueLoadPersStrmInfoFixup, pThis));
	CHKiRet(obj.Deserialize(&pThis->tVars.disk.pRead, (uchar*) "strm", psQIF,
			       (rsRetVal(*)(obj_t*,void*))qqueueLoadPersStrmInfoFixup, pThis));

	CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pWrite));
	CHKiRet(strm.SeekCurrOffs(pThis->tVars.disk.pRead));

	/* OK, we could successfully read the file, so we now can request that it be
	 * deleted when we are done with the persisted information.
	 */
	pThis->bNeedDelQIF = 1;

finalize_it:
	if(psQIF != NULL)
		strm.Destruct(&psQIF);

	if(iRet != RS_RET_OK) {
		dbgoprint((obj_t*) pThis, "error %d reading .qi file - can not read persisted info (if any)\n",
			  iRet);
	}

	RETiRet;
}


/* disk queue constructor.
 * Note that we use a file limit of 10,000,000 files. That number should never pose a
 * problem. If so, I guess the user has a design issue... But of course, the code can
 * always be changed (though it would probably be more appropriate to increase the
 * allowed file size at this point - that should be a config setting...
 * rgerhards, 2008-01-10
 */
static rsRetVal qConstructDisk(qqueue_t *pThis)
{
	DEFiRet;
	int bRestarted = 0;

	ASSERT(pThis != NULL);

	/* and now check if there is some persistent information that needs to be read in */
	iRet = qqueueTryLoadPersistedInfo(pThis);
	if(iRet == RS_RET_OK)
		bRestarted = 1;
	else if(iRet != RS_RET_FILE_NOT_FOUND)
			FINALIZE;

	if(bRestarted == 1) {
		;
	} else {
		CHKiRet(strm.Construct(&pThis->tVars.disk.pWrite));
		CHKiRet(strm.SetbSync(pThis->tVars.disk.pWrite, pThis->bSyncQueueFiles));
		CHKiRet(strm.SetDir(pThis->tVars.disk.pWrite, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir())));
		CHKiRet(strm.SetiMaxFiles(pThis->tVars.disk.pWrite, 10000000));
		CHKiRet(strm.SettOperationsMode(pThis->tVars.disk.pWrite, STREAMMODE_WRITE));
		CHKiRet(strm.SetsType(pThis->tVars.disk.pWrite, STREAMTYPE_FILE_CIRCULAR));
		CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pWrite));

		CHKiRet(strm.Construct(&pThis->tVars.disk.pRead));
		CHKiRet(strm.SetbSync(pThis->tVars.disk.pRead, pThis->bSyncQueueFiles));
		CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 1));
		CHKiRet(strm.SetDir(pThis->tVars.disk.pRead, glbl.GetWorkDir(), strlen((char*)glbl.GetWorkDir())));
		CHKiRet(strm.SetiMaxFiles(pThis->tVars.disk.pRead, 10000000));
		CHKiRet(strm.SettOperationsMode(pThis->tVars.disk.pRead, STREAMMODE_READ));
		CHKiRet(strm.SetsType(pThis->tVars.disk.pRead, STREAMTYPE_FILE_CIRCULAR));
		CHKiRet(strm.ConstructFinalize(pThis->tVars.disk.pRead));


		CHKiRet(strm.SetFName(pThis->tVars.disk.pWrite, pThis->pszFilePrefix, pThis->lenFilePrefix));
		CHKiRet(strm.SetFName(pThis->tVars.disk.pRead,  pThis->pszFilePrefix, pThis->lenFilePrefix));
	}

	/* now we set (and overwrite in case of a persisted restart) some parameters which
	 * should always reflect the current configuration variables. Be careful by doing so,
	 * for example file name generation must not be changed as that would break the
	 * ability to read existing queue files. -- rgerhards, 2008-01-12
	 */
	CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pWrite, pThis->iMaxFileSize));
	CHKiRet(strm.SetiMaxFileSize(pThis->tVars.disk.pRead, pThis->iMaxFileSize));

finalize_it:
	RETiRet;
}


static rsRetVal qDestructDisk(qqueue_t *pThis)
{
	DEFiRet;
	
	ASSERT(pThis != NULL);

	if (pThis->tVars.disk.pWrite != NULL)
		strm.Destruct(&pThis->tVars.disk.pWrite);
	if (pThis->tVars.disk.pRead != NULL)
		strm.Destruct(&pThis->tVars.disk.pRead);

	RETiRet;
}

static rsRetVal qAddDisk(qqueue_t *pThis, void* pUsr)
{
	DEFiRet;
	number_t nWriteCount;

	ASSERT(pThis != NULL);

	CHKiRet(strm.SetWCntr(pThis->tVars.disk.pWrite, &nWriteCount));
	CHKiRet((objSerialize(pUsr))(pUsr, pThis->tVars.disk.pWrite));
	CHKiRet(strm.Flush(pThis->tVars.disk.pWrite));
	CHKiRet(strm.SetWCntr(pThis->tVars.disk.pWrite, NULL)); /* no more counting for now... */

	pThis->tVars.disk.sizeOnDisk += nWriteCount;

	/* we have enqueued the user element to disk. So we now need to destruct
	 * the in-memory representation. The instance will be re-created upon
	 * dequeue. -- rgerhards, 2008-07-09
	 */
	objDestruct(pUsr);

	dbgoprint((obj_t*) pThis, "write wrote %lld octets to disk, queue disk size now %lld octets\n",
		   nWriteCount, pThis->tVars.disk.sizeOnDisk);

finalize_it:
	RETiRet;
}

static rsRetVal qDelDisk(qqueue_t *pThis, void **ppUsr)
{
	DEFiRet;

	int64 offsIn;
	int64 offsOut;

	CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pRead, &offsIn));
	CHKiRet(obj.Deserialize(ppUsr, (uchar*) "msg", pThis->tVars.disk.pRead, NULL, NULL));
	CHKiRet(strm.GetCurrOffset(pThis->tVars.disk.pRead, &offsOut));

	/* This time it is a bit tricky: we free disk space only upon file deletion. So we need
	 * to keep track of what we have read until we get an out-offset that is lower than the
	 * in-offset (which indicates file change). Then, we can subtract the whole thing from
	 * the on-disk size. -- rgerhards, 2008-01-30
	 */
	if(offsIn < offsOut) {
		pThis->tVars.disk.bytesRead += offsOut - offsIn;
	} else {
		pThis->tVars.disk.sizeOnDisk -= pThis->tVars.disk.bytesRead;
		pThis->tVars.disk.bytesRead = offsOut;
		dbgoprint((obj_t*) pThis, "a file has been deleted, now %lld octets disk space used\n", pThis->tVars.disk.sizeOnDisk);
		/* awake possibly waiting enq process */
		pthread_cond_signal(&pThis->notFull); /* we hold the mutex while we are in here! */
	}

finalize_it:
	RETiRet;
}

/* -------------------- direct (no queueing) -------------------- */
static rsRetVal qConstructDirect(qqueue_t __attribute__((unused)) *pThis)
{
	return RS_RET_OK;
}


static rsRetVal qDestructDirect(qqueue_t __attribute__((unused)) *pThis)
{
	return RS_RET_OK;
}

static rsRetVal qAddDirect(qqueue_t *pThis, void* pUsr)
{
	DEFiRet;

	ASSERT(pThis != NULL);

	/* calling the consumer is quite different here than it is from a worker thread */
	/* we need to provide the consumer's return value back to the caller because in direct
	 * mode the consumer probably has a lot to convey (which get's lost in the other modes
	 * because they are asynchronous. But direct mode is deliberately synchronous.
	 * rgerhards, 2008-02-12
	 */
	iRet = pThis->pConsumer(pThis->pUsr, pUsr);

	RETiRet;
}

static rsRetVal qDelDirect(qqueue_t __attribute__((unused)) *pThis, __attribute__((unused)) void **out)
{
	return RS_RET_OK;
}


/* --------------- end type-specific handlers -------------------- */


/* unget a user pointer that has been dequeued. This functionality is especially important
 * for consumer cancel cleanup handlers. To support it, a short list of ungotten user pointers
 * is maintened in memory.
 * rgerhards, 2008-01-20
 */
static rsRetVal
qqueueUngetObj(qqueue_t *pThis, obj_t *pUsr, int bLockMutex)
{
	DEFiRet;
	DEFVARS_mutexProtection;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ISOBJ_assert(pUsr); /* TODO: we aborted right at this place at least 3 times -- race? 2008-02-28, -03-10, -03-15
			       The second time I noticed it the queue was in destruction with NO worker threads
			       running. The pUsr ptr was totally off and provided no clue what it may be pointing
			       at (except that it looked like the static data pool). Both times, the abort happend
			       inside an action queue */

	dbgoprint((obj_t*) pThis, "ungetting user object %s\n", obj.GetName(pUsr));
	BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, bLockMutex);
	iRet = qqueueAddLinkedList(&pThis->pUngetRoot, &pThis->pUngetLast, pUsr);
	++pThis->iUngottenObjs;	/* indicate one more */
	END_MTX_PROTECTED_OPERATIONS(pThis->mut);

	RETiRet;
}


/* dequeues a user pointer from the ungotten queue. Pointers from there should always be
 * dequeued first.
 *
 * This function must only be called when the mutex is locked!
 *
 * rgerhards, 2008-01-29
 */
static rsRetVal
qqueueGetUngottenObj(qqueue_t *pThis, obj_t **ppUsr)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ASSERT(ppUsr != NULL);

	iRet = qqueueDelLinkedList(&pThis->pUngetRoot, &pThis->pUngetLast, ppUsr);
	--pThis->iUngottenObjs;	/* indicate one less */
	dbgoprint((obj_t*) pThis, "dequeued ungotten user object %s\n", obj.GetName(*ppUsr));

	RETiRet;
}


/* generic code to add a queue entry
 * We use some specific code to most efficiently support direct mode
 * queues. This is justified in spite of the gain and the need to do some
 * things truely different. -- rgerhards, 2008-02-12
 */
static rsRetVal
qqueueAdd(qqueue_t *pThis, void *pUsr)
{
	DEFiRet;

	ASSERT(pThis != NULL);

	CHKiRet(pThis->qAdd(pThis, pUsr));

	if(pThis->qType != QUEUETYPE_DIRECT) {
		ATOMIC_INC(pThis->iQueueSize);
		dbgoprint((obj_t*) pThis, "entry added, size now %d entries\n", pThis->iQueueSize);
	}

finalize_it:
	RETiRet;
}


/* generic code to remove a queue entry
 * rgerhards, 2008-01-29: we must first see if there is any object in the
 * ungotten list and, if so, dequeue it first.
 */
static rsRetVal
qqueueDel(qqueue_t *pThis, void *pUsr)
{
	DEFiRet;

	ASSERT(pThis != NULL);

	/* we do NOT abort if we encounter an error, because otherwise the queue
	 * will not be decremented, what will most probably result in an endless loop.
	 * If we decrement, however, we may lose a message. But that is better than
	 * losing the whole process because it loops... -- rgerhards, 2008-01-03
	 */
	if(pThis->iUngottenObjs > 0) {
		iRet = qqueueGetUngottenObj(pThis, (obj_t**) pUsr);
	} else {
		iRet = pThis->qDel(pThis, pUsr);
		ATOMIC_DEC(pThis->iQueueSize);
	}

	dbgoprint((obj_t*) pThis, "entry deleted, state %d, size now %d entries\n",
		  iRet, pThis->iQueueSize);

	RETiRet;
}


/* This function shuts down all worker threads and waits until they
 * have terminated. If they timeout, they are cancelled. Parameters have been set
 * before this function is called so that DA queues will be fully persisted to
 * disk (if configured to do so).
 * rgerhards, 2008-01-24
 * Please note that this function shuts down BOTH the parent AND the child queue
 * in DA case. This is necessary because their timeouts are tightly coupled. Most
 * importantly, the timeouts would be applied twice (or logic be extremely
 * complex) if each would have its own shutdown. The function does not self check
 * this condition - the caller must make sure it is not called with a parent.
 */
static rsRetVal qqueueShutdownWorkers(qqueue_t *pThis)
{
	DEFiRet;
	DEFVARS_mutexProtection;
	struct timespec tTimeout;
	rsRetVal iRetLocal;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ASSERT(pThis->pqParent == NULL); /* detect invalid calling sequence */

	dbgoprint((obj_t*) pThis, "initiating worker thread shutdown sequence\n");

	/* we reduce the low water mark in any case. This is not absolutely necessary, but
	 * it is useful because we enable DA mode at several spots below and so we do not need
	 * to think about the low water mark each time. 
	 */
	pThis->iHighWtrMrk = 1; /* if we do not do this, the DA queue will not stop! */
	pThis->iLowWtrMrk = 0;

	/* first try to shutdown the queue within the regular shutdown period */
	BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX);	/* some workers may be running in parallel! */
	if(qqueueGetOverallQueueSize(pThis) > 0) {
		if(pThis->bRunsDA) {
			/* We may have waited on the low water mark. As it may have changed, we
			 * see if we reactivate the worker.
			 */
			wtpAdviseMaxWorkers(pThis->pWtpDA, 1);
		}
	}
	END_MTX_PROTECTED_OPERATIONS(pThis->mut);

	/* Now wait for the queue's workers to shut down. Note that we run into the code even if we just found
	 * out there are no active workers - that doesn't matter: the wtp knows about that and so will
	 * return immediately.
	 * We do not yet care about the DA worker - that will be handled down later in the process.
	 * Note that we must not request shutdown right now - that may introduce a race: if the regular queue
	 * still runs DA assisted and the DA worker gets scheduled first, it will terminate itself (if the DA
	 * queue happens to be empty at that instant). Then the regular worker enqueues messages, what will lead
	 * to a restart of the worker. Of course, everything will continue to run, but in a bit sub-optimal way
	 * (from a performance point of view). So we don't do anything right now. The DA queue will continue to
	 * process messages and shutdown itself in any case if there is nothing to do. So we don't loose anything
	 * by not requesting shutdown now.
	 * rgerhards, 2008-01-25
	 */
	/* first calculate absolute timeout - we need the absolute value here, because we need to coordinate
	 * shutdown of both the regular and DA queue on *the same* timeout.
	 */
	timeoutComp(&tTimeout, pThis->toQShutdown);
	dbgoprint((obj_t*) pThis, "trying shutdown of regular workers\n");
	iRetLocal = wtpShutdownAll(pThis->pWtpReg, wtpState_SHUTDOWN, &tTimeout);
	if(iRetLocal == RS_RET_TIMED_OUT) {
		dbgoprint((obj_t*) pThis, "regular shutdown timed out on primary queue (this is OK)\n");
	} else {
		/* OK, the regular queue is now shut down. So we can now wait for the DA queue (if running DA) */
		dbgoprint((obj_t*) pThis, "regular queue workers shut down.\n");
		BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX);	/* some workers may be running in parallel! */
		if(pThis->bRunsDA) {
			END_MTX_PROTECTED_OPERATIONS(pThis->mut);
			dbgoprint((obj_t*) pThis, "we have a DA queue (0x%lx), requesting its shutdown.\n",
				 qqueueGetID(pThis->pqDA));
			/* we use the same absolute timeout as above, so we do not use more than the configured
			 * timeout interval!
			 */
			dbgoprint((obj_t*) pThis, "trying shutdown of DA workers\n");
			iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN, &tTimeout);
			if(iRetLocal == RS_RET_TIMED_OUT) {
				dbgoprint((obj_t*) pThis, "shutdown timed out on DA queue (this is OK)\n");
			}
		} else {
			END_MTX_PROTECTED_OPERATIONS(pThis->mut);
		}
	}

	/* when we reach this point, both queues are either empty or the regular queue shutdown timeout
	 * has expired. Now we need to check if we are configured to not loose messages. If so, we need
	 * to persist the queue to disk (this is only possible if the queue is DA-enabled). We must also
	 * set the primary queue to SHUTDOWN_IMMEDIATE, as it shall now terminate as soon as its consumer
	 * is done. This is especially important as we otherwise may interfere with queue order while the
	 * DA consumer is running. -- rgerhards, 2008-01-27
	 * Note: there was a note that we should not wait eternally on the DA worker if we run in
	 * enqueue-only note. I have reviewed the code and think there is no need for this check. Howerver,
	 * I'd like to keep this note in here should we happen to run into some related trouble.
	 * rgerhards, 2008-01-28
	 */
	wtpSetState(pThis->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE); /* set primary queue to shutdown only */

	/* at this stage, we need to have the DA worker properly initialized and running (if there is one) */
	if(pThis->bRunsDA)
		qqueueWaitDAModeInitialized(pThis);

	BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX);	/* some workers may be running in parallel! */
	/* optimize parameters for shutdown of DA-enabled queues */
	if(pThis->bIsDA && qqueueGetOverallQueueSize(pThis) > 0 && pThis->bSaveOnShutdown) {
		/* switch to enqueue-only mode so that no more actions happen */
		if(pThis->bRunsDA == 0) {
			qqueueInitDA(pThis, QUEUE_MODE_ENQONLY, MUTEX_ALREADY_LOCKED); /* switch to DA mode */
		} else {
			/* TODO: RACE: we may reach this point when the DA worker has been initialized (state 1)
			 * but is not yet running (state 2). In this case, pThis->pqDA is NULL! rgerhards, 2008-02-27
			 */
			qqueueSetEnqOnly(pThis->pqDA, QUEUE_MODE_ENQONLY, MUTEX_ALREADY_LOCKED); /* switch to enqueue-only mode */
		}
		END_MTX_PROTECTED_OPERATIONS(pThis->mut);
		/* make sure we do not timeout before we are done */
		dbgoprint((obj_t*) pThis, "bSaveOnShutdown configured, eternal timeout set\n");
		timeoutComp(&tTimeout, QUEUE_TIMEOUT_ETERNAL);
		/* and run the primary queue's DA worker to drain the queue */
		iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN, &tTimeout);
		if(iRetLocal != RS_RET_OK) {
			dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying to shut down primary queue in disk save mode, "
				  "continuing, but results are unpredictable\n", iRetLocal);
		}
	} else {
		END_MTX_PROTECTED_OPERATIONS(pThis->mut);
	}

	/* now the primary queue is either empty, persisted to disk - or set to loose messages. So we
	 * can now request immediate shutdown of any remaining workers. Note that if bSaveOnShutdown was set,
	 * the queue is now empty. If regular workers are still running, and try to pull the next message,
	 * they will automatically terminate as there no longer is any message left to process.
	 */
	BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX);	/* some workers may be running in parallel! */
	if(qqueueGetOverallQueueSize(pThis) > 0) {
		timeoutComp(&tTimeout, pThis->toActShutdown);
		if(wtpGetCurNumWrkr(pThis->pWtpReg, LOCK_MUTEX) > 0) {
			END_MTX_PROTECTED_OPERATIONS(pThis->mut);
			dbgoprint((obj_t*) pThis, "trying immediate shutdown of regular workers\n");
			iRetLocal = wtpShutdownAll(pThis->pWtpReg, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout);
			if(iRetLocal == RS_RET_TIMED_OUT) {
				dbgoprint((obj_t*) pThis, "immediate shutdown timed out on primary queue (this is acceptable and "
					  "triggers cancellation)\n");
			} else if(iRetLocal != RS_RET_OK) {
				dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the primary queue "
					  "in disk save mode. Continuing, but results are unpredictable\n", iRetLocal);
			}
			/* we need to re-aquire the mutex for the next check in this case! */
			BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, LOCK_MUTEX);	/* some workers may be running in parallel! */
		}
		if(pThis->bRunsDA && wtpGetCurNumWrkr(pThis->pWtpDA, LOCK_MUTEX) > 0) {
			/* and now the same for the DA queue */
			END_MTX_PROTECTED_OPERATIONS(pThis->mut);
			dbgoprint((obj_t*) pThis, "trying immediate shutdown of DA workers\n");
			iRetLocal = wtpShutdownAll(pThis->pWtpDA, wtpState_SHUTDOWN_IMMEDIATE, &tTimeout);
			if(iRetLocal == RS_RET_TIMED_OUT) {
				dbgoprint((obj_t*) pThis, "immediate shutdown timed out on DA queue (this is acceptable and "
					  "triggers cancellation)\n");
			} else if(iRetLocal != RS_RET_OK) {
				dbgoprint((obj_t*) pThis, "unexpected iRet state %d after trying immediate shutdown of the DA queue "
					  "in disk save mode. Continuing, but results are unpredictable\n", iRetLocal);
			}
		} else {
			END_MTX_PROTECTED_OPERATIONS(pThis->mut);
		}
	} else {
		END_MTX_PROTECTED_OPERATIONS(pThis->mut);
	}

	/* Now queue workers should have terminated. If not, we need to cancel them as we have applied
	 * all timeout setting. If any worker in any queue still executes, its consumer is possibly
	 * long-running and cancelling is the only way to get rid of it. Note that the
	 * cancellation handler will probably re-queue a user pointer, so the queue's enqueue
	 * function is still needed (what is no problem as we do not yet destroy the queue - but I
	 * thought it's a good idea to mention that fact). -- rgerhards, 2008-01-25
	 */
	dbgoprint((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the primary queue\n");
	iRetLocal = wtpCancelAll(pThis->pWtpReg); /* returns immediately if all threads already have terminated */
	if(iRetLocal != RS_RET_OK) {
		dbgoprint((obj_t*) pThis, "unexpected iRet state %d trying to cancel primary queue worker "
			  "threads, continuing, but results are unpredictable\n", iRetLocal);
	}


	/* TODO: think: do we really need to do this here? Can't it happen on DA queue destruction? If we 
	 * disable it, we get an assertion... I think this is OK, as we need to have a certain order and
	 * canceling the DA workers here ensures that order. But in any instant, we may have a look at this
	 * code after we have reaced the milestone. -- rgerhards, 2008-01-27
	 */
	/* ... and now the DA queue, if it exists (should always be after the primary one) */
	if(pThis->pqDA != NULL) {
		dbgoprint((obj_t*) pThis, "checking to see if we need to cancel any worker threads of the DA queue\n");
		iRetLocal = wtpCancelAll(pThis->pqDA->pWtpReg); /* returns immediately if all threads already have terminated */
		if(iRetLocal != RS_RET_OK) {
			dbgoprint((obj_t*) pThis, "unexpected iRet state %d trying to cancel DA queue worker "
				  "threads, continuing, but results are unpredictable\n", iRetLocal);
		}
	}

	/* ... finally ... all worker threads have terminated :-)
	 * Well, more precisely, they *are in termination*. Some cancel cleanup handlers
	 * may still be running. 
	 */
	dbgoprint((obj_t*) pThis, "worker threads terminated, remaining queue size %d.\n", qqueueGetOverallQueueSize(pThis));

	RETiRet;
}



/* Constructor for the queue object
 * This constructs the data structure, but does not yet start the queue. That
 * is done by queueStart(). The reason is that we want to give the caller a chance
 * to modify some parameters before the queue is actually started.
 */
rsRetVal qqueueConstruct(qqueue_t **ppThis, queueType_t qType, int iWorkerThreads,
		        int iMaxQueueSize, rsRetVal (*pConsumer)(void*,void*))
{
	DEFiRet;
	qqueue_t *pThis;

	ASSERT(ppThis != NULL);
	ASSERT(pConsumer != NULL);
	ASSERT(iWorkerThreads >= 0);

	if((pThis = (qqueue_t *)calloc(1, sizeof(qqueue_t))) == NULL) {
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);
	}

	/* we have an object, so let's fill the properties */
	objConstructSetObjInfo(pThis);
	if((pThis->pszSpoolDir = (uchar*) strdup((char*)glbl.GetWorkDir())) == NULL)
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);

	/* set some water marks so that we have useful defaults if none are set specifically */
	pThis->iFullDlyMrk  = iMaxQueueSize - (iMaxQueueSize / 100) *  3; /* default 97% */
	pThis->iLightDlyMrk = iMaxQueueSize - (iMaxQueueSize / 100) * 30; /* default 70% */

	pThis->lenSpoolDir = strlen((char*)pThis->pszSpoolDir);
	pThis->iMaxFileSize = 1024 * 1024; /* default is 1 MiB */
	pThis->iQueueSize = 0;
	pThis->iMaxQueueSize = iMaxQueueSize;
	pThis->pConsumer = pConsumer;
	pThis->iNumWorkerThreads = iWorkerThreads;
	pThis->iDeqtWinToHr = 25; /* disable time-windowed dequeuing by default */

	pThis->pszFilePrefix = NULL;
	pThis->qType = qType;

	/* set type-specific handlers and other very type-specific things (we can not totally hide it...) */
	switch(qType) {
		case QUEUETYPE_FIXED_ARRAY:
			pThis->qConstruct = qConstructFixedArray;
			pThis->qDestruct = qDestructFixedArray;
			pThis->qAdd = qAddFixedArray;
			pThis->qDel = qDelFixedArray;
			break;
		case QUEUETYPE_LINKEDLIST:
			pThis->qConstruct = qConstructLinkedList;
			pThis->qDestruct = qDestructLinkedList;
			pThis->qAdd = qAddLinkedList;
			pThis->qDel = (rsRetVal (*)(qqueue_t*,void**)) qDelLinkedList;
			break;
		case QUEUETYPE_DISK:
			pThis->qConstruct = qConstructDisk;
			pThis->qDestruct = qDestructDisk;
			pThis->qAdd = qAddDisk;
			pThis->qDel = qDelDisk;
			/* special handling */
			pThis->iNumWorkerThreads = 1; /* we need exactly one worker */
			break;
		case QUEUETYPE_DIRECT:
			pThis->qConstruct = qConstructDirect;
			pThis->qDestruct = qDestructDirect;
			pThis->qAdd = qAddDirect;
			pThis->qDel = qDelDirect;
			break;
	}

finalize_it:
	OBJCONSTRUCT_CHECK_SUCCESS_AND_CLEANUP
	RETiRet;
}


/* cancellation cleanup handler for queueWorker ()
 * Updates admin structure and frees ressources.
 * Params:
 * arg1 - user pointer (in this case a qqueue_t)
 * arg2 - user data pointer (in this case a queue data element, any object [queue's pUsr ptr!])
 * Note that arg2 may be NULL, in which case no dequeued but unprocessed pUsr exists!
 * rgerhards, 2008-01-16
 */
static rsRetVal
qqueueConsumerCancelCleanup(void *arg1, void *arg2)
{
	DEFiRet;

	qqueue_t *pThis = (qqueue_t*) arg1;
	obj_t *pUsr = (obj_t*) arg2;

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pUsr != NULL) {
		/* make sure the data element is not lost */
		dbgoprint((obj_t*) pThis, "cancelation cleanup handler consumer called, we need to unget one user data element\n");
		CHKiRet(qqueueUngetObj(pThis, pUsr, LOCK_MUTEX));
	}
	
finalize_it:
	RETiRet;
}



/* This function checks if the provided message shall be discarded and does so, if needed.
 * In DA mode, we do not discard any messages as we assume the disk subsystem is fast enough to
 * provide real-time creation of spool files.
 * Note: cached copies of iQueueSize and bRunsDA are provided so that no mutex locks are required.
 * The caller must have obtained them while the mutex was locked. Of course, these values may no
 * longer be current, but that is OK for the discard check. At worst, the message is either processed
 * or discarded when it should not have been. As discarding is in itself somewhat racy and erratic,
 * that is no problems for us. This function MUST NOT lock the queue mutex, it could result in
 * deadlocks!
 * If the message is discarded, it can no longer be processed by the caller. So be sure to check
 * the return state!
 * rgerhards, 2008-01-24
 */
static int qqueueChkDiscardMsg(qqueue_t *pThis, int iQueueSize, int bRunsDA, void *pUsr)
{
	DEFiRet;
	rsRetVal iRetLocal;
	int iSeverity;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ISOBJ_assert(pUsr);

	if(pThis->iDiscardMrk > 0 && iQueueSize >= pThis->iDiscardMrk && bRunsDA == 0) {
		iRetLocal = objGetSeverity(pUsr, &iSeverity);
		if(iRetLocal == RS_RET_OK && iSeverity >= pThis->iDiscardSeverity) {
			dbgoprint((obj_t*) pThis, "queue nearly full (%d entries), discarded severity %d message\n",
				  iQueueSize, iSeverity);
			objDestruct(pUsr);
			ABORT_FINALIZE(RS_RET_QUEUE_FULL);
		} else {
			dbgoprint((obj_t*) pThis, "queue nearly full (%d entries), but could not drop msg "
				  "(iRet: %d, severity %d)\n", iQueueSize, iRetLocal, iSeverity);
		}
	}

finalize_it:
	RETiRet;
}


/* dequeue the queued object for the queue consumers.
 * rgerhards, 2008-10-21
 */
static rsRetVal
qqueueDequeueConsumable(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave)
{
	DEFiRet;
	void *pUsr;
	int iQueueSize;
	int bRunsDA;	 /* cache for early mutex release */

	/* dequeue element (still protected from mutex) */
	iRet = qqueueDel(pThis, &pUsr);
	qqueueChkPersist(pThis);
	iQueueSize = qqueueGetOverallQueueSize(pThis); /* cache this for after mutex release */
	bRunsDA = pThis->bRunsDA; /* cache this for after mutex release */

	/* We now need to save the user pointer for the cancel cleanup handler, BUT ONLY
	 * if we could successfully obtain a user pointer. Otherwise, we would bring the
	 * cancel cleanup handler into big troubles (and we did ;)). Note that we can
	 * NOT set the variable further below, as this may lead to an object leak. We 
	 * may get cancelled before we reach that part of the code, so the only 
	 * solution is to do it here. -- rgerhards, 2008-02-27
	 */
	if(iRet == RS_RET_OK) {
		pWti->pUsrp = pUsr;
	}

	/* awake some flow-controlled sources if we can do this right now */
	/* TODO: this could be done better from a performance point of view -- do it only if
	 * we have someone waiting for the condition (or only when we hit the watermark right
	 * on the nail [exact value]) -- rgerhards, 2008-03-14
	 */
	if(iQueueSize < pThis->iFullDlyMrk / 2) {
		pthread_cond_broadcast(&pThis->belowFullDlyWtrMrk);
	}

	if(iQueueSize < pThis->iLightDlyMrk / 2) {
		pthread_cond_broadcast(&pThis->belowLightDlyWtrMrk);
	}

	/* rgerhards, 2008-09-30: I reversed the order of cond_signal und mutex_unlock
	 * as of the pthreads recommendation on predictable scheduling behaviour. I don't see
	 * any problems caused by this, but I add this comment in case some will be seen
	 * in the next time.
	 */
	pthread_cond_signal(&pThis->notFull);
	d_pthread_mutex_unlock(pThis->mut);
	pthread_setcancelstate(iCancelStateSave, NULL);
	/* WE ARE NO LONGER PROTECTED BY THE MUTEX */

	/* do actual processing (the lengthy part, runs in parallel)
	 * If we had a problem while dequeing, we do not call the consumer,
	 * but we otherwise ignore it. This is in the hopes that it will be
	 * self-healing. However, this is really not a good thing.
	 * rgerhards, 2008-01-03
	 */
	if(iRet != RS_RET_OK)
		FINALIZE;

	/* we are running in normal, non-disk-assisted mode do a quick check if we need to drain the queue.
	 * In DA mode, we do not discard any messages as we assume the disk subsystem is fast enough to
	 * provide real-time creation of spool files.
	 * Note: It is OK to use the cached iQueueSize here, because it does not hurt if it is slightly wrong.
	 */
	CHKiRet(qqueueChkDiscardMsg(pThis, iQueueSize, bRunsDA, pUsr));

finalize_it:
	if(iRet != RS_RET_OK && iRet != RS_RET_DISCARDMSG) {
		dbgoprint((obj_t*) pThis, "error %d dequeueing element - ignoring, but strange things "
			  "may happen\n", iRet);
	}
	RETiRet;
}


/* The rate limiter
 *
 * Here we may wait if a dequeue time window is defined or if we are
 * rate-limited. TODO: If we do so, we should also look into the
 * way new worker threads are spawned. Obviously, it doesn't make much
 * sense to spawn additional worker threads when none of them can do any
 * processing. However, it is deemed acceptable to allow this for an initial
 * implementation of the timeframe/rate limiting feature.
 * Please also note that these feature could also be implemented at the action
 * level. However, that would limit them to be used together with actions. We have
 * taken the broader approach, moving it right into the queue. This is even
 * necessary if we want to prevent spawning of multiple unnecessary worker
 * threads as described above. -- rgerhards, 2008-04-02
 *
 *
 * time window: tCurr is current time; tFrom is start time, tTo is end time (in mil 24h format).
 * We may have tFrom = 4, tTo = 10 --> run from 4 to 10 hrs. nice and happy
 * we may also have tFrom= 22, tTo = 4 -> run from 10pm to 4am, which is actually two
 *     windows: 0-4; 22-23:59
 * so when to run? Let's assume we have 3am
 *
 * if(tTo < tFrom) {
 * 	if(tCurr < tTo [3 < 4] || tCurr > tFrom [3 > 22])
 * 		do work
 * 	else
 * 		sleep for tFrom - tCurr "hours" [22 - 5 --> 17]
 * } else {
 * 	if(tCurr >= tFrom [3 >= 4] && tCurr < tTo [3 < 10])
 * 		do work
 * 	else
 * 		sleep for tTo - tCurr "hours" [4 - 3 --> 1]
 * }
 *
 * Bottom line: we need to check which type of window we have and need to adjust our
 * logic accordingly. Of course, sleep calculations need to be done up to the minute, 
 * but you get the idea from the code above.
 */
static rsRetVal
qqueueRateLimiter(qqueue_t *pThis)
{
	DEFiRet;
	int iDelay;
	int iHrCurr;
	time_t tCurr;
	struct tm m;

	ISOBJ_TYPE_assert(pThis, qqueue);

	iDelay = 0;
	if(pThis->iDeqtWinToHr != 25) { /* 25 means disabled */
		/* time calls are expensive, so only do them when needed */
		time(&tCurr);
		localtime_r(&tCurr, &m);
		iHrCurr = m.tm_hour;

		if(pThis->iDeqtWinToHr < pThis->iDeqtWinFromHr) {
			if(iHrCurr < pThis->iDeqtWinToHr || iHrCurr > pThis->iDeqtWinFromHr) {
				; /* do not delay */
			} else {
				iDelay = (pThis->iDeqtWinFromHr - iHrCurr) * 3600;
				/* this time, we are already into the next hour, so we need
				 * to subtract our current minute and seconds.
				 */
				iDelay -= m.tm_min * 60;
				iDelay -= m.tm_sec;
			}
		} else {
			if(iHrCurr >= pThis->iDeqtWinFromHr && iHrCurr < pThis->iDeqtWinToHr) {
				; /* do not delay */
			} else {
				if(iHrCurr < pThis->iDeqtWinFromHr) {
					iDelay = (pThis->iDeqtWinFromHr - iHrCurr - 1) * 3600; /* -1 as we are already in the hour */
					iDelay += (60 - m.tm_min) * 60;
					iDelay += 60 - m.tm_sec;
				} else {
					iDelay = (24 - iHrCurr + pThis->iDeqtWinFromHr) * 3600;
					/* this time, we are already into the next hour, so we need
					 * to subtract our current minute and seconds.
					 */
					iDelay -= m.tm_min * 60;
					iDelay -= m.tm_sec;
				}
			}
		}
	}

	if(iDelay > 0) {
		dbgoprint((obj_t*) pThis, "outside dequeue time window, delaying %d seconds\n", iDelay);
		srSleep(iDelay, 0);
	}

	RETiRet;
}



/* This is the queue consumer in the regular (non-DA) case. It is 
 * protected by the queue mutex, but MUST release it as soon as possible.
 * rgerhards, 2008-01-21
 */
static rsRetVal
qqueueConsumerReg(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ISOBJ_TYPE_assert(pWti, wti);

	CHKiRet(qqueueDequeueConsumable(pThis, pWti, iCancelStateSave));
	CHKiRet(pThis->pConsumer(pThis->pUsr, pWti->pUsrp));

	/* we now need to check if we should deliberately delay processing a bit
	 * and, if so, do that. -- rgerhards, 2008-01-30
	 */
	if(pThis->iDeqSlowdown) {
		dbgoprint((obj_t*) pThis, "sleeping %d microseconds as requested by config params\n",
			  pThis->iDeqSlowdown);
		srSleep(pThis->iDeqSlowdown / 1000000, pThis->iDeqSlowdown % 1000000);
	}

finalize_it:
	RETiRet;
}


/* This is a special consumer to feed the disk-queue in disk-assited mode.
 * When active, our own queue more or less acts as a memory buffer to the disk.
 * So this consumer just needs to drain the memory queue and submit entries
 * to the disk queue. The disk queue will then call the actual consumer from
 * the app point of view (we chain two queues here).
 * When this method is entered, the mutex is always locked and needs to be unlocked
 * as part of the processing.
 * rgerhards, 2008-01-14
 */
static rsRetVal
qqueueConsumerDA(qqueue_t *pThis, wti_t *pWti, int iCancelStateSave)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ISOBJ_TYPE_assert(pWti, wti);

	CHKiRet(qqueueDequeueConsumable(pThis, pWti, iCancelStateSave));
	CHKiRet(qqueueEnqObj(pThis->pqDA, eFLOWCTL_NO_DELAY, pWti->pUsrp));

finalize_it:
	dbgoprint((obj_t*) pThis, "DAConsumer returns with iRet %d\n", iRet);
	RETiRet;
}


/* must only be called when the queue mutex is locked, else results
 * are not stable!
 * If we are a child, we have done our duty when the queue is empty. In that case,
 * we can terminate.
 * Version for the DA worker thread. NOTE: the pThis->bRunsDA is different from
 * the DA queue
 */
static int
qqueueChkStopWrkrDA(qqueue_t *pThis)
{
	/* if our queue is in destruction, we drain to the DA queue and so we shall not terminate
	 * until we have done so.
	 */
	int bStopWrkr;

	BEGINfunc

	if(pThis->bEnqOnly) {
		bStopWrkr = 1;
	} else {
		if(pThis->bRunsDA == 2) {
			ASSERT(pThis->pqDA != NULL);
			if(   pThis->pqDA->bEnqOnly
			   && pThis->pqDA->sizeOnDiskMax > 0
			   && pThis->pqDA->tVars.disk.sizeOnDisk > pThis->pqDA->sizeOnDiskMax) {
				/* this queue can never grow, so we can give up... */
				bStopWrkr = 1;
			} else if(qqueueGetOverallQueueSize(pThis) < pThis->iHighWtrMrk && pThis->bQueueStarted == 1) {
				bStopWrkr = 1;
			} else {
				bStopWrkr = 0;
			}
		} else {
			bStopWrkr = 1;
		}
	}

	ENDfunc
	return  bStopWrkr;
}


/* must only be called when the queue mutex is locked, else results
 * are not stable!
 * If we are a child, we have done our duty when the queue is empty. In that case,
 * we can terminate.
 * Version for the regular worker thread. NOTE: the pThis->bRunsDA is different from
 * the DA queue
 */
static int
qqueueChkStopWrkrReg(qqueue_t *pThis)
{
	return pThis->bEnqOnly || pThis->bRunsDA || (pThis->pqParent != NULL && qqueueGetOverallQueueSize(pThis) == 0);
}


/* must only be called when the queue mutex is locked, else results
 * are not stable! DA queue version
 */
static int
qqueueIsIdleDA(qqueue_t *pThis)
{
	/* remember: iQueueSize is the DA queue size, not the main queue! */
	/* TODO: I think we need just a single function for DA and non-DA mode - but I leave it for now as is */
	return(qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk));
}
/* must only be called when the queue mutex is locked, else results
 * are not stable! Regular queue version
 */
static int
qqueueIsIdleReg(qqueue_t *pThis)
{
#if 0 /* enable for performance testing */
	int ret;
	ret = qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk);
	if(ret) fprintf(stderr, "queue is idle\n");
	return ret;
#else 
	/* regular code! */
	return(qqueueGetOverallQueueSize(pThis) == 0 || (pThis->bRunsDA && qqueueGetOverallQueueSize(pThis) <= pThis->iLowWtrMrk));
#endif
}


/* This function is called when a worker thread for the regular queue is shut down.
 * If we are the primary queue, this is not really interesting to us. If, however,
 * we are the DA (child) queue, that means the DA queue is empty. In that case, we
 * need to signal the parent queue's DA worker, so that it can terminate DA mode.
 * rgerhards, 2008-01-26
 * rgerhards, 2008-02-27: HOWEVER, in a shutdown condition, it may be that the parent's worker thread pool
 * has already been terminated and destructed. This *is* a legal condition and happens
 * from time to time in practice. So we need to signal only if there still is a
 * parent DA worker queue. Please keep in mind that the the parent's DA worker
 * pool is DIFFERENT from our (DA queue) regular worker pool. So when the parent's
 * pWtpDA is destructed, there can still be some of our (DAq/wtp) threads be running.
 * I am telling this, because I, too, always get confused by those...
 */
static rsRetVal
qqueueRegOnWrkrShutdown(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pThis->pqParent != NULL) {
		pThis->pqParent->bChildIsDone = 1; /* indicate we are done */
		if(pThis->pqParent->pWtpDA != NULL) { /* see comment in function header from 2008-02-27 */
			wtpAdviseMaxWorkers(pThis->pqParent->pWtpDA, 1); /* reactivate DA worker (always 1) */
		}
	}

	RETiRet;
}


/* The following function is called when a regular queue worker starts up. We need this
 * hook to indicate in the parent queue (if we are a child) that we are not done yet.
 */
static rsRetVal
qqueueRegOnWrkrStartup(qqueue_t *pThis)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pThis->pqParent != NULL) {
		pThis->pqParent->bChildIsDone = 0;
	}

	RETiRet;
}


/* start up the queue - it must have been constructed and parameters defined
 * before.
 */
rsRetVal qqueueStart(qqueue_t *pThis) /* this is the ConstructionFinalizer */
{
	DEFiRet;
	rsRetVal iRetLocal;
	int bInitialized = 0; /* is queue already initialized? */
	uchar pszBuf[64];
	size_t lenBuf;

	ASSERT(pThis != NULL);

	/* we need to do a quick check if our water marks are set plausible. If not,
	 * we correct the most important shortcomings. TODO: do that!!!! -- rgerhards, 2008-03-14
	 */

	/* finalize some initializations that could not yet be done because it is
	 * influenced by properties which might have been set after queueConstruct ()
	 */
	if(pThis->pqParent == NULL) {
		pThis->mut = (pthread_mutex_t *) malloc (sizeof (pthread_mutex_t));
		pthread_mutex_init(pThis->mut, NULL);
	} else {
		/* child queue, we need to use parent's mutex */
		dbgoprint((obj_t*) pThis, "I am a child\n");
		pThis->mut = pThis->pqParent->mut;
	}

	pthread_mutex_init(&pThis->mutThrdMgmt, NULL);
	pthread_cond_init (&pThis->condDAReady, NULL);
	pthread_cond_init (&pThis->notFull, NULL);
	pthread_cond_init (&pThis->notEmpty, NULL);
	pthread_cond_init (&pThis->belowFullDlyWtrMrk, NULL);
	pthread_cond_init (&pThis->belowLightDlyWtrMrk, NULL);

	/* call type-specific constructor */
	CHKiRet(pThis->qConstruct(pThis)); /* this also sets bIsDA */

	dbgoprint((obj_t*) pThis, "type %d, enq-only %d, disk assisted %d, maxFileSz %lld, qsize %d, child %d, "
				  "full delay %d, light delay %d starting\n",
		  pThis->qType, pThis->bEnqOnly, pThis->bIsDA, pThis->iMaxFileSize,
		  qqueueGetOverallQueueSize(pThis), pThis->pqParent == NULL ? 0 : 1,
		  pThis->iFullDlyMrk, pThis->iLightDlyMrk);

	if(pThis->qType == QUEUETYPE_DIRECT)
		FINALIZE;	/* with direct queues, we are already finished... */

	/* create worker thread pools for regular operation. The DA pool is created on an as-needed
	 * basis, which potentially means never under most circumstances.
	 */
	lenBuf = snprintf((char*)pszBuf, sizeof(pszBuf), "%s:Reg", obj.GetName((obj_t*) pThis));
	CHKiRet(wtpConstruct		(&pThis->pWtpReg));
	CHKiRet(wtpSetDbgHdr		(pThis->pWtpReg, pszBuf, lenBuf));
	CHKiRet(wtpSetpfRateLimiter	(pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRateLimiter));
	CHKiRet(wtpSetpfChkStopWrkr	(pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int)) qqueueChkStopWrkrReg));
	CHKiRet(wtpSetpfIsIdle		(pThis->pWtpReg, (rsRetVal (*)(void *pUsr, int)) qqueueIsIdleReg));
	CHKiRet(wtpSetpfDoWork		(pThis->pWtpReg, (rsRetVal (*)(void *pUsr, void *pWti, int)) qqueueConsumerReg));
	CHKiRet(wtpSetpfOnWorkerCancel	(pThis->pWtpReg, (rsRetVal (*)(void *pUsr, void*pWti))qqueueConsumerCancelCleanup));
	CHKiRet(wtpSetpfOnWorkerStartup	(pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRegOnWrkrStartup));
	CHKiRet(wtpSetpfOnWorkerShutdown(pThis->pWtpReg, (rsRetVal (*)(void *pUsr)) qqueueRegOnWrkrShutdown));
	CHKiRet(wtpSetpmutUsr		(pThis->pWtpReg, pThis->mut));
	CHKiRet(wtpSetpcondBusy		(pThis->pWtpReg, &pThis->notEmpty));
	CHKiRet(wtpSetiNumWorkerThreads	(pThis->pWtpReg, pThis->iNumWorkerThreads));
	CHKiRet(wtpSettoWrkShutdown	(pThis->pWtpReg, pThis->toWrkShutdown));
	CHKiRet(wtpSetpUsr		(pThis->pWtpReg, pThis));
	CHKiRet(wtpConstructFinalize	(pThis->pWtpReg));

	/* initialize worker thread instances */
	if(pThis->bIsDA) {
		/* If we are disk-assisted, we need to check if there is a QIF file
		 * which we need to load. -- rgerhards, 2008-01-15
		 */
		iRetLocal = qqueueHaveQIF(pThis);
		if(iRetLocal == RS_RET_OK) {
			dbgoprint((obj_t*) pThis, "on-disk queue present, needs to be reloaded\n");
			qqueueInitDA(pThis, QUEUE_MODE_ENQDEQ, LOCK_MUTEX); /* initiate DA mode */
			bInitialized = 1; /* we are done */
		} else {
			/* TODO: use logerror? -- rgerhards, 2008-01-16 */
			dbgoprint((obj_t*) pThis, "error %d trying to access on-disk queue files, starting without them. "
			          "Some data may be lost\n", iRetLocal);
		}
	}

	if(!bInitialized) {
		dbgoprint((obj_t*) pThis, "queue starts up without (loading) any DA disk state (this is normal for the DA "
			  "queue itself!)\n");
	}

	/* if the queue already contains data, we need to start the correct number of worker threads. This can be
	 * the case when a disk queue has been loaded. If we did not start it here, it would never start.
	 */
	qqueueAdviseMaxWorkers(pThis);
	pThis->bQueueStarted = 1;

finalize_it:
	RETiRet;
}


/* persist the queue to disk. If we have something to persist, we first
 * save the information on the queue properties itself and then we call
 * the queue-type specific drivers.
 * Variable bIsCheckpoint is set to 1 if the persist is for a checkpoint,
 * and 0 otherwise.
 * rgerhards, 2008-01-10
 */
static rsRetVal qqueuePersist(qqueue_t *pThis, int bIsCheckpoint)
{
	DEFiRet;
	strm_t *psQIF = NULL; /* Queue Info File */
	uchar pszQIFNam[MAXFNAME];
	size_t lenQIFNam;
	obj_t *pUsr;

	ASSERT(pThis != NULL);

	if(pThis->qType != QUEUETYPE_DISK) {
		if(qqueueGetOverallQueueSize(pThis) > 0) {
			/* This error code is OK, but we will probably not implement this any time
 			 * The reason is that persistence happens via DA queues. But I would like to
			 * leave the code as is, as we so have a hook in case we need one.
			 * -- rgerhards, 2008-01-28
			 */
			ABORT_FINALIZE(RS_RET_NOT_IMPLEMENTED);
		} else
			FINALIZE; /* if the queue is empty, we are happy and done... */
	}

	dbgoprint((obj_t*) pThis, "persisting queue to disk, %d entries...\n", qqueueGetOverallQueueSize(pThis));

	/* Construct file name */
	lenQIFNam = snprintf((char*)pszQIFNam, sizeof(pszQIFNam) / sizeof(uchar), "%s/%s.qi",
			     (char*) glbl.GetWorkDir(), (char*)pThis->pszFilePrefix);

	if((bIsCheckpoint != QUEUE_CHECKPOINT) && (qqueueGetOverallQueueSize(pThis) == 0)) {
		if(pThis->bNeedDelQIF) {
			unlink((char*)pszQIFNam);
			pThis->bNeedDelQIF = 0;
		}
		/* indicate spool file needs to be deleted */
		if (pThis->tVars.disk.pRead != NULL)
			CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 1));
		FINALIZE; /* nothing left to do, so be happy */
	}

	CHKiRet(strm.Construct(&psQIF));
	CHKiRet(strm.SettOperationsMode(psQIF, STREAMMODE_WRITE_TRUNC));
	CHKiRet(strm.SetbSync(psQIF, pThis->bSyncQueueFiles));
	CHKiRet(strm.SetsType(psQIF, STREAMTYPE_FILE_SINGLE));
	CHKiRet(strm.SetFName(psQIF, pszQIFNam, lenQIFNam));
	CHKiRet(strm.ConstructFinalize(psQIF));

	/* first, write the property bag for ourselfs
	 * And, surprisingly enough, we currently need to persist only the size of the
	 * queue. All the rest is re-created with then-current config parameters when the
	 * queue is re-created. Well, we'll also save the current queue type, just so that
	 * we know when somebody has changed the queue type... -- rgerhards, 2008-01-11
	 */
	CHKiRet(obj.BeginSerializePropBag(psQIF, (obj_t*) pThis));
	objSerializeSCALAR(psQIF, iQueueSize, INT);
	objSerializeSCALAR(psQIF, iUngottenObjs, INT);
	objSerializeSCALAR(psQIF, tVars.disk.sizeOnDisk, INT64);
	objSerializeSCALAR(psQIF, tVars.disk.bytesRead, INT64);
	CHKiRet(obj.EndSerialize(psQIF));

	/* now we must persist all objects on the ungotten queue - they can not go to
	 * to the regular files. -- rgerhards, 2008-01-29
	 */
	while(pThis->iUngottenObjs > 0) {
		CHKiRet(qqueueGetUngottenObj(pThis, &pUsr));
		CHKiRet((objSerialize(pUsr))(pUsr, psQIF));
		objDestruct(pUsr);
	}

	/* now persist the stream info */
	if (pThis->tVars.disk.pWrite != NULL)
		CHKiRet(strm.Serialize(pThis->tVars.disk.pWrite, psQIF));
	if (pThis->tVars.disk.pRead != NULL)
		CHKiRet(strm.Serialize(pThis->tVars.disk.pRead, psQIF));
	
	/* tell the input file object that it must not delete the file on close if the queue
	 * is non-empty - but only if we are not during a simple checkpoint
	 */
	if(bIsCheckpoint != QUEUE_CHECKPOINT && pThis->tVars.disk.pRead != NULL) {
		CHKiRet(strm.SetbDeleteOnClose(pThis->tVars.disk.pRead, 0));
	}

	/* we have persisted the queue object. So whenever it comes to an empty queue,
	 * we need to delete the QIF. Thus, we indicte that need.
	 */
	pThis->bNeedDelQIF = 1;

finalize_it:
	if(psQIF != NULL)
		strm.Destruct(&psQIF);

	RETiRet;
}


/* check if we need to persist the current queue info. If an
 * error occurs, thus should be ignored by caller (but we still
 * abide to our regular call interface)...
 * rgerhards, 2008-01-13
 */
static rsRetVal qqueueChkPersist(qqueue_t *pThis)
{
	ISOBJ_TYPE_assert(pThis, qqueue);

	if(pThis->iPersistUpdCnt && ++pThis->iUpdsSincePersist >= pThis->iPersistUpdCnt) {
		qqueuePersist(pThis, QUEUE_CHECKPOINT);
		pThis->iUpdsSincePersist = 0;
	}

	return RS_RET_OK;
}


/* destructor for the queue object */
BEGINobjDestruct(qqueue) /* be sure to specify the object type also in END and CODESTART macros! */
CODESTARTobjDestruct(qqueue)
	pThis->bQueueInDestruction = 1; /* indicate we are in destruction (modifies some behaviour) */

	/* shut down all workers (handles *all* of the persistence logic)
	 * See function head comment of queueShutdownWorkers () on why we don't call it
	 * We also do not need to shutdown workers when we are in enqueue-only mode or we are a
	 * direct queue - because in both cases we have none... ;)
	 * with a child! -- rgerhards, 2008-01-28
	 */
	if(pThis->qType != QUEUETYPE_DIRECT && !pThis->bEnqOnly && pThis->pqParent == NULL)
		qqueueShutdownWorkers(pThis);

	/* finally destruct our (regular) worker thread pool
	 * Note: currently pWtpReg is never NULL, but if we optimize our logic, this may happen,
	 * e.g. when they are not created in enqueue-only mode. We already check the condition
	 * as this may otherwise be very hard to find once we optimize (and have long forgotten
	 * about this condition here ;)
	 * rgerhards, 2008-01-25
	 */
	if(pThis->qType != QUEUETYPE_DIRECT && pThis->pWtpReg != NULL) {
		wtpDestruct(&pThis->pWtpReg);
	}

	/* Now check if we actually have a DA queue and, if so, destruct it.
	 * Note that the wtp must be destructed first, it may be in cancel cleanup handler
	 * *right now* and actually *need* to access the queue object to persist some final
	 * data (re-queueing case). So we need to destruct the wtp first, which will make 
	 * sure all workers have terminated. Please note that this also generates a situation
	 * where it is possible that the DA queue has a parent pointer but the parent has
	 * no WtpDA associated with it - which is perfectly legal thanks to this code here.
	 */
	if(pThis->pWtpDA != NULL) {
		wtpDestruct(&pThis->pWtpDA);
	}
	if(pThis->pqDA != NULL) {
		qqueueDestruct(&pThis->pqDA);
	}

	/* persist the queue (we always do that - queuePersits() does cleanup if the queue is empty)
	 * This handler is most important for disk queues, it will finally persist the necessary
	 * on-disk structures. In theory, other queueing modes may implement their other (non-DA)
	 * methods of persisting a queue between runs, but in practice all of this is done via
	 * disk queues and DA mode. Anyhow, it doesn't hurt to know that we could extend it here
	 * if need arises (what I doubt...) -- rgerhards, 2008-01-25
	 */
	CHKiRet_Hdlr(qqueuePersist(pThis, QUEUE_NO_CHECKPOINT)) {
		dbgoprint((obj_t*) pThis, "error %d persisting queue - data lost!\n", iRet);
	}

	/* finally, clean up some simple things... */
	if(pThis->pqParent == NULL) {
		/* if we are not a child, we allocated our own mutex, which we now need to destroy */
		pthread_mutex_destroy(pThis->mut);
		free(pThis->mut);
	}
	pthread_mutex_destroy(&pThis->mutThrdMgmt);
	pthread_cond_destroy(&pThis->condDAReady);
	pthread_cond_destroy(&pThis->notFull);
	pthread_cond_destroy(&pThis->notEmpty);
	pthread_cond_destroy(&pThis->belowFullDlyWtrMrk);
	pthread_cond_destroy(&pThis->belowLightDlyWtrMrk);

	/* type-specific destructor */
	iRet = pThis->qDestruct(pThis);

	if(pThis->pszFilePrefix != NULL)
		free(pThis->pszFilePrefix);

	if(pThis->pszSpoolDir != NULL)
		free(pThis->pszSpoolDir);
ENDobjDestruct(qqueue)


/* set the queue's file prefix
 * The passed-in string is duplicated. So if the caller does not need
 * it any longer, it must free it.
 * rgerhards, 2008-01-09
 */
rsRetVal
qqueueSetFilePrefix(qqueue_t *pThis, uchar *pszPrefix, size_t iLenPrefix)
{
	DEFiRet;

	if(pThis->pszFilePrefix != NULL)
		free(pThis->pszFilePrefix);

	if(pszPrefix == NULL) /* just unset the prefix! */
		ABORT_FINALIZE(RS_RET_OK);

	if((pThis->pszFilePrefix = malloc(sizeof(uchar) * iLenPrefix + 1)) == NULL)
		ABORT_FINALIZE(RS_RET_OUT_OF_MEMORY);
	memcpy(pThis->pszFilePrefix, pszPrefix, iLenPrefix + 1);
	pThis->lenFilePrefix = iLenPrefix;

finalize_it:
	RETiRet;
}

/* set the queue's maximum file size
 * rgerhards, 2008-01-09
 */
rsRetVal
qqueueSetMaxFileSize(qqueue_t *pThis, size_t iMaxFileSize)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	
	if(iMaxFileSize < 1024) {
		ABORT_FINALIZE(RS_RET_VALUE_TOO_LOW);
	}

	pThis->iMaxFileSize = iMaxFileSize;

finalize_it:
	RETiRet;
}


/* enqueue a new user data element
 * Enqueues the new element and awakes worker thread.
 */
rsRetVal
qqueueEnqObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr)
{
	DEFiRet;
	int iCancelStateSave;
	struct timespec t;

	ISOBJ_TYPE_assert(pThis, qqueue);

	/* first check if we need to discard this message (which will cause CHKiRet() to exit)
	 * rgerhards, 2008-10-07: It is OK to do this outside of mutex protection. The iQueueSize
	 * and bRunsDA parameters may not reflect the correct settings here, but they are
	 * "good enough" in the sense that they can be used to drive the decision. Valgrind's
	 * threading tools may point this access to be an error, but this is done
	 * intentional. I do not see this causes problems to us.
	 */
	CHKiRet(qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pThis->bRunsDA, pUsr));

	/* Please note that this function is not cancel-safe and consequently
	 * sets the calling thread's cancelibility state to PTHREAD_CANCEL_DISABLE
	 * during its execution. If that is not done, race conditions occur if the
	 * thread is canceled (most important use case is input module termination).
	 * rgerhards, 2008-01-08
	 */
	if(pThis->qType != QUEUETYPE_DIRECT) {
		pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave);
		d_pthread_mutex_lock(pThis->mut);
	}

	/* then check if we need to add an assistance disk queue */
	if(pThis->bIsDA)
		CHKiRet(qqueueChkStrtDA(pThis));
	
	/* handle flow control
	 * There are two different flow control mechanisms: basic and advanced flow control.
	 * Basic flow control has always been implemented and protects the queue structures
	 * in that it makes sure no more data is enqueued than the queue is configured to
	 * support. Enhanced flow control is being added today. There are some sources which
	 * can easily be stopped, e.g. a file reader. This is the case because it is unlikely
	 * that blocking those sources will have negative effects (after all, the file is
	 * continued to be written). Other sources can somewhat be blocked (e.g. the kernel
	 * log reader or the local log stream reader): in general, nothing is lost if messages
	 * from these sources are not picked up immediately. HOWEVER, they can not block for
	 * an extended period of time, as this either causes message loss or - even worse - some
	 * other bad effects (e.g. unresponsive system in respect to the main system log socket).
	 * Finally, there are some (few) sources which can not be blocked at all. UDP syslog is
	 * a prime example. If a UDP message is not received, it is simply lost. So we can't
	 * do anything against UDP sockets that come in too fast. The core idea of advanced
	 * flow control is that we take into account the different natures of the sources and
	 * select flow control mechanisms that fit these needs. This also means, in the end
	 * result, that non-blockable sources like UDP syslog receive priority in the system.
	 * It's a side effect, but a good one ;) -- rgerhards, 2008-03-14
	 */
	if(flowCtlType == eFLOWCTL_FULL_DELAY) {
		while(pThis->iQueueSize >= pThis->iFullDlyMrk) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: FullDelay mark reached for full delayable message - blocking.\n");
			pthread_cond_wait(&pThis->belowFullDlyWtrMrk, pThis->mut); /* TODO error check? But what do then? */
		}
	} else if(flowCtlType == eFLOWCTL_LIGHT_DELAY) {
		if(pThis->iQueueSize >= pThis->iLightDlyMrk) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: LightDelay mark reached for light delayable message - blocking a bit.\n");
			timeoutComp(&t, 1000); /* 1000 millisconds = 1 second TODO: make configurable */
			pthread_cond_timedwait(&pThis->belowLightDlyWtrMrk, pThis->mut, &t); /* TODO error check? But what do then? */
		}
	}

	/* from our regular flow control settings, we are now ready to enqueue the object.
	 * However, we now need to do a check if the queue permits to add more data. If that
	 * is not the case, basic flow control enters the field, which means we wait for
	 * the queue to become ready or drop the new message. -- rgerhards, 2008-03-14
	 */
	while(   (pThis->iMaxQueueSize > 0 && pThis->iQueueSize >= pThis->iMaxQueueSize)
	      || (pThis->qType == QUEUETYPE_DISK && pThis->sizeOnDiskMax != 0
	      	  && pThis->tVars.disk.sizeOnDisk > pThis->sizeOnDiskMax)) {
		dbgoprint((obj_t*) pThis, "enqueueMsg: queue FULL - waiting to drain.\n");
		timeoutComp(&t, pThis->toEnq);
		if(pthread_cond_timedwait(&pThis->notFull, pThis->mut, &t) != 0) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n");
			objDestruct(pUsr);
			ABORT_FINALIZE(RS_RET_QUEUE_FULL);
		}
	}

	/* and finally enqueue the message */
	CHKiRet(qqueueAdd(pThis, pUsr));
	qqueueChkPersist(pThis);

finalize_it:
	if(pThis->qType != QUEUETYPE_DIRECT) {
		/* make sure at least one worker is running. */
		qqueueAdviseMaxWorkers(pThis);
		/* and release the mutex */
		d_pthread_mutex_unlock(pThis->mut);
		pthread_setcancelstate(iCancelStateSave, NULL);
		dbgoprint((obj_t*) pThis, "EnqueueMsg advised worker start\n");
	}

	RETiRet;
}


/* enqueue a single data object. This currently is a helper to qqueueMultiEnqObj.
 * Note that the queue mutex MUST already be locked when this function is called.
 * rgerhards, 2009-06-16
 */
static inline rsRetVal
doEnqSingleObj(qqueue_t *pThis, flowControl_t flowCtlType, void *pUsr)
{
	DEFiRet;
	struct timespec t;

	/* first check if we need to discard this message (which will cause CHKiRet() to exit)
	 */
	CHKiRet(qqueueChkDiscardMsg(pThis, pThis->iQueueSize, pThis->bRunsDA, pUsr));

	/* then check if we need to add an assistance disk queue */
	if(pThis->bIsDA)
		CHKiRet(qqueueChkStrtDA(pThis));
	
	/* handle flow control
	 * There are two different flow control mechanisms: basic and advanced flow control.
	 * Basic flow control has always been implemented and protects the queue structures
	 * in that it makes sure no more data is enqueued than the queue is configured to
	 * support. Enhanced flow control is being added today. There are some sources which
	 * can easily be stopped, e.g. a file reader. This is the case because it is unlikely
	 * that blocking those sources will have negative effects (after all, the file is
	 * continued to be written). Other sources can somewhat be blocked (e.g. the kernel
	 * log reader or the local log stream reader): in general, nothing is lost if messages
	 * from these sources are not picked up immediately. HOWEVER, they can not block for
	 * an extended period of time, as this either causes message loss or - even worse - some
	 * other bad effects (e.g. unresponsive system in respect to the main system log socket).
	 * Finally, there are some (few) sources which can not be blocked at all. UDP syslog is
	 * a prime example. If a UDP message is not received, it is simply lost. So we can't
	 * do anything against UDP sockets that come in too fast. The core idea of advanced
	 * flow control is that we take into account the different natures of the sources and
	 * select flow control mechanisms that fit these needs. This also means, in the end
	 * result, that non-blockable sources like UDP syslog receive priority in the system.
	 * It's a side effect, but a good one ;) -- rgerhards, 2008-03-14
	 */
	if(flowCtlType == eFLOWCTL_FULL_DELAY) {
		while(pThis->iQueueSize >= pThis->iFullDlyMrk) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: FullDelay mark reached for full delayable message - blocking.\n");
			pthread_cond_wait(&pThis->belowFullDlyWtrMrk, pThis->mut); /* TODO error check? But what do then? */
		}
	} else if(flowCtlType == eFLOWCTL_LIGHT_DELAY) {
		if(pThis->iQueueSize >= pThis->iLightDlyMrk) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: LightDelay mark reached for light delayable message - blocking a bit.\n");
			timeoutComp(&t, 1000); /* 1000 millisconds = 1 second TODO: make configurable */
			pthread_cond_timedwait(&pThis->belowLightDlyWtrMrk, pThis->mut, &t); /* TODO error check? But what do then? */
		}
	}

	/* from our regular flow control settings, we are now ready to enqueue the object.
	 * However, we now need to do a check if the queue permits to add more data. If that
	 * is not the case, basic flow control enters the field, which means we wait for
	 * the queue to become ready or drop the new message. -- rgerhards, 2008-03-14
	 */
	while(   (pThis->iMaxQueueSize > 0 && pThis->iQueueSize >= pThis->iMaxQueueSize)
	      || (pThis->qType == QUEUETYPE_DISK && pThis->sizeOnDiskMax != 0
	      	  && pThis->tVars.disk.sizeOnDisk > pThis->sizeOnDiskMax)) {
		dbgoprint((obj_t*) pThis, "enqueueMsg: queue FULL - waiting to drain.\n");
		timeoutComp(&t, pThis->toEnq);
		if(pthread_cond_timedwait(&pThis->notFull, pThis->mut, &t) != 0) {
			dbgoprint((obj_t*) pThis, "enqueueMsg: cond timeout, dropping message!\n");
			objDestruct(pUsr);
			ABORT_FINALIZE(RS_RET_QUEUE_FULL);
		}
	}

	/* and finally enqueue the message */
	CHKiRet(qqueueAdd(pThis, pUsr));
	qqueueChkPersist(pThis); // TODO: optimize, do in outer function! (but we need parts from v5?)

finalize_it:
	RETiRet;
}

/* enqueue multiple user data elements at once. The aim is to provide a faster interface
 * for object submission. Uses the multi_submit_t helper object.
 * Please note that this function is not cancel-safe and consequently
 * sets the calling thread's cancelibility state to PTHREAD_CANCEL_DISABLE
 * during its execution. If that is not done, race conditions occur if the
 * thread is canceled (most important use case is input module termination).
 * rgerhards, 2009-06-16
 */
rsRetVal
qqueueMultiEnqObj(qqueue_t *pThis, multi_submit_t *pMultiSub)
{
	int iCancelStateSave;
	int i;
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	assert(pMultiSub != NULL);

	if(pThis->qType != QUEUETYPE_DIRECT) {
		pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &iCancelStateSave);
		d_pthread_mutex_lock(pThis->mut);
	}

	for(i = 0 ; i < pMultiSub->nElem ; ++i) {
dbgprintf("queueMultiEnq: %d\n", i);
		CHKiRet(doEnqSingleObj(pThis, pMultiSub->ppMsgs[i]->flowCtlType, (void*)pMultiSub->ppMsgs[i]));
	}

finalize_it:
	if(pThis->qType != QUEUETYPE_DIRECT) {
		/* make sure at least one worker is running. */
		qqueueAdviseMaxWorkers(pThis);
		/* and release the mutex */
		d_pthread_mutex_unlock(pThis->mut);
		pthread_setcancelstate(iCancelStateSave, NULL);
		dbgoprint((obj_t*) pThis, "MultiEnqObj advised worker start\n");
	}

	RETiRet;
}


/* set queue mode to enqueue only or not
 * There is one subtle issue: this method may be called during queue
 * construction or while it is running. In the former case, the queue 
 * mutex does not yet exist (it is NULL), while in the later case it
 * must be locked. The function detects the state and operates as 
 * required.
 * rgerhards, 2008-01-16
 */
static rsRetVal
qqueueSetEnqOnly(qqueue_t *pThis, int bEnqOnly, int bLockMutex)
{
	DEFiRet;
	DEFVARS_mutexProtection;

	ISOBJ_TYPE_assert(pThis, qqueue);

	/* for simplicity, we do one big mutex lock. This method is extremely seldom
	 * called, so that doesn't matter... -- rgerhards, 2008-01-16
	 */
	if(pThis->mut != NULL) {
		BEGIN_MTX_PROTECTED_OPERATIONS(pThis->mut, bLockMutex);
	}

	if(bEnqOnly == pThis->bEnqOnly)
		FINALIZE; /* no change, nothing to do */

	if(pThis->bQueueStarted) {
		/* we need to adjust queue operation only if we are not during initial param setup */
		if(bEnqOnly == 1) {
			/* switch to enqueue-only mode */
			/* this means we need to terminate all workers - that's it... */
			dbgoprint((obj_t*) pThis, "switching to enqueue-only mode, terminating all worker threads\n");
			if(pThis->pWtpReg != NULL)
				wtpWakeupAllWrkr(pThis->pWtpReg);
			if(pThis->pWtpDA != NULL)
				wtpWakeupAllWrkr(pThis->pWtpDA);
		} else {
			/* switch back to regular mode */
			ABORT_FINALIZE(RS_RET_NOT_IMPLEMENTED); /* we don't need this so far... */
		}
	}

	pThis->bEnqOnly = bEnqOnly;

finalize_it:
	if(pThis->mut != NULL) {
		END_MTX_PROTECTED_OPERATIONS(pThis->mut);
	}
	RETiRet;
}


/* some simple object access methods */
DEFpropSetMeth(qqueue, bSyncQueueFiles, int)
DEFpropSetMeth(qqueue, iPersistUpdCnt, int)
DEFpropSetMeth(qqueue, iDeqtWinFromHr, int)
DEFpropSetMeth(qqueue, iDeqtWinToHr, int)
DEFpropSetMeth(qqueue, toQShutdown, long)
DEFpropSetMeth(qqueue, toActShutdown, long)
DEFpropSetMeth(qqueue, toWrkShutdown, long)
DEFpropSetMeth(qqueue, toEnq, long)
DEFpropSetMeth(qqueue, iHighWtrMrk, int)
DEFpropSetMeth(qqueue, iLowWtrMrk, int)
DEFpropSetMeth(qqueue, iDiscardMrk, int)
DEFpropSetMeth(qqueue, iFullDlyMrk, int)
DEFpropSetMeth(qqueue, iDiscardSeverity, int)
DEFpropSetMeth(qqueue, bIsDA, int)
DEFpropSetMeth(qqueue, iMinMsgsPerWrkr, int)
DEFpropSetMeth(qqueue, bSaveOnShutdown, int)
DEFpropSetMeth(qqueue, pUsr, void*)
DEFpropSetMeth(qqueue, iDeqSlowdown, int)
DEFpropSetMeth(qqueue, sizeOnDiskMax, int64)


/* This function can be used as a generic way to set properties. Only the subset
 * of properties required to read persisted property bags is supported. This
 * functions shall only be called by the property bag reader, thus it is static.
 * rgerhards, 2008-01-11
 */
#define isProp(name) !rsCStrSzStrCmp(pProp->pcsName, (uchar*) name, sizeof(name) - 1)
static rsRetVal qqueueSetProperty(qqueue_t *pThis, var_t *pProp)
{
	DEFiRet;

	ISOBJ_TYPE_assert(pThis, qqueue);
	ASSERT(pProp != NULL);

 	if(isProp("iQueueSize")) {
		pThis->iQueueSize = pProp->val.num;
 	} else if(isProp("iUngottenObjs")) {
		pThis->iUngottenObjs = pProp->val.num;
 	} else if(isProp("tVars.disk.sizeOnDisk")) {
		pThis->tVars.disk.sizeOnDisk = pProp->val.num;
 	} else if(isProp("tVars.disk.bytesRead")) {
		pThis->tVars.disk.bytesRead = pProp->val.num;
 	} else if(isProp("qType")) {
		if(pThis->qType != pProp->val.num)
			ABORT_FINALIZE(RS_RET_QTYPE_MISMATCH);
	}

finalize_it:
	RETiRet;
}
#undef	isProp

/* dummy */
rsRetVal qqueueQueryInterface(void) { return RS_RET_NOT_IMPLEMENTED; }

/* Initialize the stream class. Must be called as the very first method
 * before anything else is called inside this class.
 * rgerhards, 2008-01-09
 */
BEGINObjClassInit(qqueue, 1, OBJ_IS_CORE_MODULE)
	/* request objects we use */
	CHKiRet(objUse(glbl, CORE_COMPONENT));
	CHKiRet(objUse(strm, CORE_COMPONENT));

	/* now set our own handlers */
	OBJSetMethodHandler(objMethod_SETPROPERTY, qqueueSetProperty);
ENDObjClassInit(qqueue)

/* vi:set ai:
 */