From da30f79c9e35ab8cca71601a33665af72d0880ff Mon Sep 17 00:00:00 2001 From: Jeff Darcy Date: Wed, 1 Feb 2017 21:54:30 -0500 Subject: glusterd: double-check whether brick is alive for stats With multiplexing, our tests detach bricks from their host processes without glusterd being involved. Thus, when we ask glusterd to fetch profile info, it will try to fetch from a brick that's actually not present any more. While it can handle the process being dead and its RPC connection being closed, it barfs if it gets a negative response from a live brick process. This is not a problem in normal use, because the brick can't disappear without glusterd seeing it. The fix is to double check that the brick is actually running, by looking for its pidfile which the tests *do* clean up as part of killing a brick. Backport of: > Change-Id: I098465b175ecf23538bd7207357c752a2bba8f4e > BUG: 1385758 > Reviewed-on: https://review.gluster.org/16509 BUG: 1418091 Change-Id: Ia61e273134520c8ccfa3371ee2370cb9a1920877 Signed-off-by: Jeff Darcy Reviewed-on: https://review.gluster.org/16532 Smoke: Gluster Build System NetBSD-regression: NetBSD Build System CentOS-regression: Gluster Build System Reviewed-by: Shyamsundar Ranganathan --- xlators/mgmt/glusterd/src/glusterd-op-sm.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.c b/xlators/mgmt/glusterd/src/glusterd-op-sm.c index d9b18e00195..6bc01f702cc 100644 --- a/xlators/mgmt/glusterd/src/glusterd-op-sm.c +++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.c @@ -6315,15 +6315,14 @@ glusterd_bricks_select_profile_volume (dict_t *dict, char **op_errstr, glusterd_brickinfo_t *brickinfo = NULL; glusterd_pending_node_t *pending_node = NULL; char *brick = NULL; - - + int32_t pid = -1; + char pidfile[PATH_MAX] = {0}; this = THIS; GF_ASSERT (this); priv = this->private; GF_ASSERT (priv); - ret = dict_get_str (dict, "volname", &volname); if (ret) { gf_msg ("glusterd", GF_LOG_ERROR, 0, @@ -6383,6 +6382,18 @@ glusterd_bricks_select_profile_volume (dict_t *dict, char **op_errstr, cds_list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) { if (glusterd_is_brick_started (brickinfo)) { + /* + * In normal use, glusterd_is_brick_started + * will give us the answer we need. However, + * in our tests the brick gets detached behind + * our back, so we need to double-check this + * way. + */ + GLUSTERD_GET_BRICK_PIDFILE (pidfile, volinfo, + brickinfo, priv); + if (!gf_is_service_running (pidfile, &pid)) { + continue; + } pending_node = GF_CALLOC (1, sizeof (*pending_node), gf_gld_mt_pending_node_t); if (!pending_node) { -- cgit