glusterfs.git, branch v3.4.0alpha2

performance/write-behind: guarantee non-overlapping concurrent writes

2013-03-07T07:30:14+00:00

Maintain a list of writes (either written behind or SYNC) which are
currently "in progress" (i.e, STACK_WIND'ed towards server) and hold
off any new STACK_WIND of write (either written behind or SYNC) which
overlaps with any of the "in progress" writes.

This is a guarantee which AFR's eager-lock depends upon (though not
strictly a write-behind requirement)

Change-Id: Icedd0b51b440366a906dc9223d62b7fd6ef2ca03
BUG: 857673
Original-author: Anand Avati 
Signed-off-by: Anand Avati 
Signed-off-by: Jeff Darcy 
Reviewed-on: http://review.gluster.org/4642
Tested-by: Gluster Build System

performance/write-behind: Add test case for fd being marked bad after write failures.

2013-03-07T06:37:22+00:00

BUG: 765473
Change-Id: Ia5d9fecc7f84ee4d51f8037e2dd1ed03f0394bd9
Signed-off-by: Raghavendra G 
Reviewed-on: http://review.gluster.org/4632
Reviewed-by: Jeff Darcy 
Tested-by: Gluster Build System

glusterd: Increasing throughput of synctask based mgmt ops.

2013-03-07T06:06:30+00:00

Change-Id: Ibd963f78707b157fc4c9729aa87206cfd5ecfe81
BUG: 913662
Signed-off-by: Krishnan Parthasarathi 
Reviewed-on: http://review.gluster.org/4638
Tested-by: Gluster Build System 
Reviewed-by: Jeff Darcy

volgen: Use bind-address option for bricks when option set on glusterd

2013-03-07T06:05:38+00:00

Brick processes listen on all the interfaces on a given port.
When multiple glusterds run on one machine, glusterd assumes
that it 'owns' the ports on that machine. This can lead to the
different glusterd instances to step on each other's ports.
This fix ensures that brick processes listen only on the its
host IP when glusterd has bind-address option set.

Change-Id: I4c1b05643c64d3098bf56e977e768e611ffce0f5
BUG: 913662
Signed-off-by: Krishnan Parthasarathi 
Reviewed-on: http://review.gluster.org/4637
Tested-by: Gluster Build System 
Reviewed-by: Jeff Darcy

synctask: support for (assymetric) counted barriers

2013-03-07T05:32:34+00:00

[Backport of Avati's patch on master - http://review.gluster.org/4558]
This patch introduces a new set of primitives:

  - synctask_barrier_init (stub)
  - synctask_barrier_waitfor (stub, count)
  - synctask_barrier_wake (stub)

Unlike pthread_barrier_t, this barrier has an explicit notion of
"waiter" and "waker". The "waiter" waits for @count number of
"wakers" to call synctask_barrier_wake() before returning. The
wait performed by the waiter via synctask_barrier_waitfor() is
co-operative in nature and yields the thread for scheduling other
synctasks in the mean time.

Intended use case:

  Eliminate excessive serialization in glusterd and allow for
concurrent RPC transactions.

  Code which are currently in this format:

---old---

  list_for_each_entry (peerinfo, peers, op_peers_list) {
          ...
          GD_SYNCOP (peerinfo->rpc, stub, rpc_cbk, ...);
  }

  ...

  int rpc_cbk (rpc, stub, ...)
  {
          ...
          __wake (stub);
  }

---old---

  Can be restructred into the format:

---new---

  synctask_barrier_init (stub);
  {
          list_for_each_entry (peerinfo, peers, op_peers_list) {
                  ...
                  rpc_submit (peerinfo->rpc, stub, rpc_cbk, ...);
                  count++;
           }
   }
   synctask_barrier_wait (stub, count);

   ...

   int rpc_cbk (rpc, stub, ...)
   {
           ...
           synctask_barrier_wake (stub);
   }

---new---

  In the above structure, from the synctask's point of view, the region
between synctask_barrier_init() and synctask_barrier_wait() are spawning
off asynchronous "threads" (or RPC) and keep count of how many such
threads have been spawned. Each of those threads are expected to make
one call to synctask_barrier_wake(). The call to synctask_barrier_wait()
makes the synctask thread co-operatively wait/sleep till @count such threads
call their wake function.

  This way, the synctask thread retains the "synchronous" flow in the code,
yet at the same time allows for asynchronous "threads" to acheive parallelism
over RPC.

Change-Id: Ie037f99b2d306b71e63e3a56353daec06fb0bf41
BUG: 913662
Signed-off-by: Anand Avati 
Signed-off-by: Krishnan Parthasarathi 
Reviewed-on: http://review.gluster.org/4636
Reviewed-by: Jeff Darcy 
Tested-by: Gluster Build System

testcase for the open-behind xlator when open fails

2013-03-07T05:31:46+00:00

Test if the fops which are put into a stub and are waiting for
the open to complete should be unwound with the error if open
call itself fails.

Change-Id: I8c363d98303a7df1a0ca9ea6ef207c7123fdd388
BUG: 846240
Original-author: Raghavendra Bhat 
Signed-off-by: Raghavendra Bhat 
Signed-off-by: Jeff Darcy 
Reviewed-on: http://review.gluster.org/4634
Tested-by: Gluster Build System

performance/write-behind: mark fd bad if any written behind writes fail

2013-03-07T05:18:00+00:00

BUG: 765473
Change-Id: I1ddd6ef9f5361aed96f97aa1344823836c6ddecb
Signed-off-by: Raghavendra G 
Reviewed-on: http://review.gluster.org/4630
Tested-by: Gluster Build System 
Reviewed-by: Jeff Darcy

tests: move common funtion definitions to include.rc

2013-03-07T05:13:08+00:00

BUG: 765473
Change-Id: Id0d194374d34cfec8ee601090f7fe38b1856ac22
Signed-off-by: Raghavendra G 
Reviewed-on: http://review.gluster.org/4631
Tested-by: Gluster Build System 
Reviewed-by: Jeff Darcy

tests/fileio.rc: library for file descriptor based IO in tests

2013-03-06T12:22:59+00:00

There are situations in test scripts where we want to keep open file
descriptors while performing other commands. Bash has abilities
to manage file descriptors by numbers, but the syntax is a little
brain damaging.

This library provides wrappers around it to abstract away bash's
syntax and also provides a helper function to pick a free file
descriptor on the fly.

The APIs are pretty self explanatory.

Change-Id: I82f1d1957646dd6c468d9e85c90ec30c978c7ad6
BUG: 764966
Signed-off-by: Raghavendra G 
Reviewed-on: http://review.gluster.org/4635
Reviewed-by: Jeff Darcy 
Tested-by: Jeff Darcy

Do not call xdr_string() with a NULL error message

2013-03-06T05:04:09+00:00

It is illegal to call xdr_string() with a NULL string. Linux
just retruns false, NetBSD gets a SIGSEGV when xdr_string()
calls strlen(NULL)

BUG: 916439
Change-Id: Ia958470ada6e8e55a86d439922ec942d038f5f13
Original-author: Emmanuel Dreyfus 
Signed-off-by: Emmanuel Dreyfus 
Signed-off-by: Jeff Darcy 
Reviewed-on: http://review.gluster.org/4629