diff options
author | Atin Mukherjee <amukherj@redhat.com> | 2016-03-31 11:01:53 +0530 |
---|---|---|
committer | Jeff Darcy <jdarcy@redhat.com> | 2016-04-01 13:38:54 -0700 |
commit | 34899d71f21fd2b4c523b68ffb2d7c655c776641 (patch) | |
tree | 4a818e42665ba441c71da92c891c3e2c2f4a7f7d /xlators/mgmt/glusterd/src/glusterd-utils.c | |
parent | d7001f3606866c1d9bb3cefd2767af56833dd53f (diff) |
glusterd: Allocate fresh port on brick (re)start
There is no point of using the same port through the entire volume life cycle
for a particular bricks process since there is no guarantee that the same port
would be free and no other application wouldn't consume it in between the
glusterd/volume restart.
We hit a race where on glusterd restart the daemon services start followed by
brick processes and the time brick process tries to bind with the port which was
allocated by glusterd before a restart is been already consumed by some other
client like NFS/SHD/...
Note : This is a short term solution as here we reduce the race window but don't
eliminate it completely. As a long term solution the port allocation has to be
done by glusterfsd and the same should be communicated back to glusterd for book
keeping
Change-Id: Ibbd1e7ca87e51a7cd9cf216b1fe58ef7783aef24
BUG: 1322805
Signed-off-by: Atin Mukherjee <amukherj@redhat.com>
Reviewed-on: http://review.gluster.org/13865
Smoke: Gluster Build System <jenkins@build.gluster.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Diffstat (limited to 'xlators/mgmt/glusterd/src/glusterd-utils.c')
-rw-r--r-- | xlators/mgmt/glusterd/src/glusterd-utils.c | 38 |
1 files changed, 32 insertions, 6 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c index a0cc5d409c8..55597cd39cf 100644 --- a/xlators/mgmt/glusterd/src/glusterd-utils.c +++ b/xlators/mgmt/glusterd/src/glusterd-utils.c @@ -1725,9 +1725,22 @@ glusterd_volume_start_glusterfs (glusterd_volinfo_t *volinfo, if (gf_is_service_running (pidfile, NULL)) goto connect; - port = brickinfo->port; - if (!port) - port = pmap_registry_alloc (THIS); + /* Do a pmap registry remove on the older connected port */ + if (brickinfo->port) { + ret = pmap_registry_remove (this, brickinfo->port, + brickinfo->path, + GF_PMAP_PORT_BRICKSERVER, + NULL); + if (ret) { + gf_msg (this->name, GF_LOG_ERROR, 0, + GD_MSG_PMAP_REMOVE_FAIL, "Failed to remove pmap" + "registry for older signin port %d", + brickinfo->port); + goto out; + } + } + + port = pmap_registry_alloc (THIS); /* Build the exp_path, before starting the glusterfsd even in valgrind mode. Otherwise all the glusterfsd processes start @@ -1790,9 +1803,22 @@ glusterd_volume_start_glusterfs (glusterd_volinfo_t *volinfo, if (volinfo->transport_type != GF_TRANSPORT_BOTH_TCP_RDMA) { runner_argprintf (&runner, "%d", port); } else { - rdma_port = brickinfo->rdma_port; - if (!rdma_port) - rdma_port = pmap_registry_alloc (THIS); + /* Do a pmap registry remove on the older connected port */ + if (brickinfo->rdma_port) { + ret = pmap_registry_remove (this, brickinfo->rdma_port, + brickinfo->path, + GF_PMAP_PORT_BRICKSERVER, + NULL); + if (ret) { + gf_msg (this->name, GF_LOG_ERROR, 0, + GD_MSG_PMAP_REMOVE_FAIL, "Failed to " + "remove pmap registry for older signin " + "port %d", brickinfo->rdma_port); + goto out; + } + } + rdma_port = pmap_registry_alloc (THIS); + runner_argprintf (&runner, "%d,%d", port, rdma_port); runner_add_arg (&runner, "--xlator-option"); runner_argprintf (&runner, "%s-server.transport.rdma.listen-port=%d", |