Acticate paxos-repair
It seems paxos-repair.timer
(nodetool repair --paxos-only
executed hourly) is needed on each node if we want system.paxos
to be clean up:
I stopped the timer early on Monday afternoon and I executed the paxos-repair.service
2 times on each node this morning.
I increased the random delay to try and prevent services being run at the same time on two nodes.
octocatalog-diff
*** Running octocatalog-diff on host cassandra01.internal.softwareheritage.org
I, [2024-12-18T14:55:29.770536 #393884] INFO -- : Catalogs compiled for cassandra01.internal.softwareheritage.org
I, [2024-12-18T14:55:29.963417 #393884] INFO -- : Diffs computed for cassandra01.internal.softwareheritage.org
diff origin/production/cassandra01.internal.softwareheritage.org current/cassandra01.internal.softwareheritage.org
*******************************************
File[/etc/systemd/system/paxos-repair.timer] =>
parameters =>
content =>
@@ -4,5 +4,5 @@
[Timer]
OnCalendar=hourly
-RandomizedDelaySec=15m
+RandomizedDelaySec=55m
FixedRandomDelay=true
Unit=paxos-repair.service
*******************************************
Service[paxos-repair.timer] =>
parameters =>
ensure =>
- false
+ true
*******************************************
Systemd::Timer[paxos-repair.timer] =>
parameters =>
active =>
- false
+ true
timer_content =>
@@ -4,5 +4,5 @@
[Timer]
OnCalendar=hourly
-RandomizedDelaySec=15m
+RandomizedDelaySec=55m
FixedRandomDelay=true
Unit=paxos-repair.service
*******************************************
Systemd::Unit_file[paxos-repair.timer] =>
parameters =>
active =>
- false
+ true
content =>
@@ -4,5 +4,5 @@
[Timer]
OnCalendar=hourly
-RandomizedDelaySec=15m
+RandomizedDelaySec=55m
FixedRandomDelay=true
Unit=paxos-repair.service
*******************************************
*** End octocatalog-diff on cassandra01.internal.softwareheritage.org