Stop flapping alerts#1582
Conversation
taurus-forever
left a comment
There was a problem hiding this comment.
@dstathis thank you for the PR! Can you please share the root cause of the issue?
Are the problem in the real outage/lack of patroni leader OR alerts are really fake/noisy?
In the first case, we should fix the stability, in the second fix alerts.
I believe 5 minutes downtime is unacceptable slow/long for 99.999% uptime target.
Anyway, defaults here defined by @7annaba3l business use cases.
AFAIK, COS have a way to fine-tune defaults per deployment,
while charm should be safe with it's default => provide maximum uptime,
otherwise immediately alert about the outage.
|
I am not really sure if there is a real problem. I haven't noticed any issues with related apps. There do seem to be some error logs but I don't really know how to interpret them. |
|
Well... is it some unstable networking environment? The The message Once K8s etcd/DCS is reachable again => the new Primary elected => all restored. We will consider to bump 0m slightly, as it might be too agressive (@7annaba3l to provide a good value), but 5m is too high for sure. We should trace your DCS instabilities and either fix the origin OR fine tune timeouts to avoid the real Primary (writes) outages. It will fix alerts instead of muting them. WDYT? @dstathis |
Increased alert duration for PatroniHasNoLeader from 0m to 5m. Signed-off-by: Dylan Stephano-Shachter <dylan.stephano-shachter@canonical.com>
Signed-off-by: Dylan Stephano-Shachter <dylan.stephano-shachter@canonical.com>
Issue
The Postgresql charm constantly has flapping alerts. I see these three regularly.
Solution
Only fire if the condition is detected for 5 minutes or more.
Checklist