diff options
author | Jeff Darcy <jdarcy@redhat.com> | 2017-01-31 14:49:45 -0500 |
---|---|---|
committer | Shyamsundar Ranganathan <srangana@redhat.com> | 2017-02-01 19:54:58 -0500 |
commit | 83803b4b2d70e9e6e16bb050d7ac8e49ba420893 (patch) | |
tree | 9a6c1f3f9a723bf578f78c624d3ce9f44baac6db /tests/volume.rc | |
parent | 80b04666ec7019e132f76f734a88559457702f1b (diff) |
core: run many bricks within one glusterfsd process
This patch adds support for multiple brick translator stacks running in
a single brick server process. This reduces our per-brick memory usage
by approximately 3x, and our appetite for TCP ports even more. It also
creates potential to avoid process/thread thrashing, and to improve QoS
by scheduling more carefully across the bricks, but realizing that
potential will require further work.
Multiplexing is controlled by the "cluster.brick-multiplex" global
option. By default it's off, and bricks are started in separate
processes as before. If multiplexing is enabled, then *compatible*
bricks (mostly those with the same transport options) will be started in
the same process.
Backport of:
> Change-Id: I45059454e51d6f4cbb29a4953359c09a408695cb
> BUG: 1385758
> Reviewed-on: https://review.gluster.org/14763
Change-Id: I4bce9080f6c93d50171823298fdf920258317ee8
BUG: 1418091
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: https://review.gluster.org/16496
Smoke: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Shyamsundar Ranganathan <srangana@redhat.com>
Diffstat (limited to 'tests/volume.rc')
-rw-r--r-- | tests/volume.rc | 30 |
1 files changed, 27 insertions, 3 deletions
diff --git a/tests/volume.rc b/tests/volume.rc index e3ae408f971..9ed92edb248 100644 --- a/tests/volume.rc +++ b/tests/volume.rc @@ -246,19 +246,43 @@ function quotad_up_status { gluster volume status | grep "Quota Daemon" | awk '{print $7}' } -function get_brick_pid { +function get_brick_pidfile { local vol=$1 local host=$2 local brick=$3 local brick_hiphenated=$(echo $brick | tr '/' '-') - echo `cat $GLUSTERD_WORKDIR/vols/$vol/run/${host}${brick_hiphenated}.pid` + echo $GLUSTERD_WORKDIR/vols/$vol/run/${host}${brick_hiphenated}.pid +} + +function get_brick_pid { + cat $(get_brick_pidfile $*) } function kill_brick { local vol=$1 local host=$2 local brick=$3 - kill -9 $(get_brick_pid $vol $host $brick) + + local pidfile=$(get_brick_pidfile $vol $host $brick) + local cmdline="/proc/$(cat $pidfile)/cmdline" + local socket=$(cat $cmdline | tr '\0' '\n' | grep '\.socket$') + + gf_attach -d $socket $brick + # Since we're not going through glusterd, we need to clean up the + # pidfile ourselves. However, other state in glusterd (e.g. + # started_here) won't be updated. A "stop-brick" CLI command would + # sure be useful. + rm -f $pidfile + + # When the last brick in a process is terminated, the process has to + # sleep for a second to give the RPC response a chance to get back to + # GlusterD. Without that, we get random failures in tests that use + # "volume stop" whenever the process termination is observed before the + # RPC response. However, that same one-second sleep can cause other + # random failures in tests that assume a brick will already be gone + # before "gf_attach -d" returns. There are too many of those to fix, + # so we compensate by putting the same one-second sleep here. + sleep 1 } function check_option_help_presence { |