summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--README7
-rw-r--r--TODO18
2 files changed, 19 insertions, 6 deletions
diff --git a/README b/README
index ab6ab77..8e18029 100644
--- a/README
+++ b/README
@@ -10,3 +10,10 @@ Running the debuginfofs server
1) vi /etc/httpd/conf.d/dav-debuginfo.conf
- Make sure the Alias line matches exportdir in config.sh
2) service httpd start
+
+Some rough stats about disk usage:
+- F10 i386+x86_64 full debuginfo stats:
+ - debuginfo data (.debug files): ~100k files, 73GB
+ - sources: ~2.2m files (!), 38GB
+- F10+Rawhide, i386+x86_64:
+ - debuginfo data: ~200k files, 111GB
diff --git a/TODO b/TODO
index da12224..46a808b 100644
--- a/TODO
+++ b/TODO
@@ -13,21 +13,21 @@
== Further work ==
Server:
+- Flag to remove out-of-date (unused) debuginfo files
+ - Rawhide builds expire very quickly
+ - Keeping rawhide debuginfo causes disk usage to grow *very* rapidly
- Cleanups for debuginfofs-mirror:
- make --verbose work properly
- Figure out if we can make yum handle checking for already-cached files
- make debuginfofs-mirror able to use rsync?
- Avahi announce server to local network
-- Optionally delete source files, since we're not using them
- - For the record: F10 i386+x86_64 full debuginfo stats:
- - debuginfo data (.debug files): 100k files, 73GB
- - source files : 2.2m files (!), 38GB
- - Saves 34% disk space, lots and lots of inodes
Client:
- Use a better FUSE-webdav implementation
- - cache files locally
- multithreaded
- yes, wdfs is single-threaded. bluh.
+ - support seek() / byte ranges
+ - wdfs does a GET on the entire file - and keeps it in RAM
+ - Optional: cache files locally
== Crazy future ideas ==
- Provide source files as well
@@ -39,3 +39,9 @@ Client:
- Clear old files from cache periodically
- Use same cache layout as static tools
- Export same exportdir layout, too
+- Specialized client-side debuginfofs
+ - Multi-threaded, read-only webdav client
+ - Cache file attrs forever
+ - Do seek()/byte ranges
+ - Optional: background-cache any file that we read
+ - Uses more bandwidth initially, less for repeated traces