Description of problem:
Currently, systemd is used to manage glusterd, but after the initial start, it does not ensure glusterd continues to run. Within limits, systemd should attempt to restart glusterd if it crashes in order to better handle transient failures.
Version-Release number of selected component (if applicable):
glusterfs-fuse-3.12.2-25.el7rhgs.x86_64
python2-gluster-3.12.2-25.el7rhgs.x86_64
gluster-nagios-common-0.2.4-1.el7rhgs.noarch
glusterfs-libs-3.12.2-25.el7rhgs.x86_64
glusterfs-client-xlators-3.12.2-25.el7rhgs.x86_64
glusterfs-cli-3.12.2-25.el7rhgs.x86_64
glusterfs-api-3.12.2-25.el7rhgs.x86_64
glusterfs-3.12.2-25.el7rhgs.x86_64
vdsm-gluster-4.19.43-2.3.el7rhgs.noarch
glusterfs-server-3.12.2-25.el7rhgs.x86_64
gluster-nagios-addons-0.2.10-2.el7rhgs.x86_64
pcp-pmda-gluster-4.3.0-0.201812061439.git24488c63.el7.x86_64
glusterfs-geo-replication-3.12.2-25.el7rhgs.x86_64
libvirt-daemon-driver-storage-gluster-4.5.0-10.el7_6.3.x86_64
glusterfs-rdma-3.12.2-25.el7rhgs.x86_64
How reproducible:
100%... if glusterd crashes, it stays down.
Steps to Reproduce:
1. Encounter glusterd SEGV
2. Observe the lack of restart
Actual results:
Glusterd is not automatically restarted on failure
Expected results:
For occasional crashes, we should use systemd to restart glusterd
Additional info:
This request comes from my experience maintaining openshift.io. We encounter periodic crashes of gd, usually due to monitoring operations. In order to have automatic recovery from these crashes, I have adjusted the unit file as follows...
In the [Service] section, I have added:
StartLimitBurst=3
StartLimitIntervalSec=3600
StartLimitInterval=3600
Restart=on-abnormal
RestartSec=60
The above causes systemd to automatically restart glusterd if it crashes. It will restart up to 3 times over a 1 hour period. This has the effect of masking the occasional failure, but will leave the daemon down if failures exceed the threshold (at which point other monitoring will raise an alert).
We should consider incorporating the above (or a variant thereof) into the standard distribution.
REVIEW: https://review.gluster.org/23751 (glusterd: start glusterd automatically on abnormal shutdown) posted (#1) for review on master by Sanju Rakonde