You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In certain cases, it is possible for two (or more) different aggregation groups to share an nflog. The conditions for this to happen are the following:
There must be at least two routes with the same matchers and group labels (group_by).
At least one of the routes must have continue: true.
This configuration meets all of the requirements to share an nflog.
When an Alertmanager is run with this configuration, you can see from the flushing alerts and Notify success lines that two aggregation groups are created and then flushed:
I expect this to create even more issues if the routes have different timers (i.e. group_wait, group_interval and repeat_interval) or active and mute time intervals:
This commit replaces the code in route.Key() with that of route.ID(),
and removes route.ID(). The motivation behind this change is to fix
a number of bugs caused by conflicting group keys such as
"Different aggregation groups can share the same nflog" (prometheus#3808)
and also prevent an issue where groups are incorrectly marked as muted
when they are not.
Signed-off-by: George Robinson <george.robinson@grafana.com>
What did you do?
In certain cases, it is possible for two (or more) different aggregation groups to share an nflog. The conditions for this to happen are the following:
group_by
).continue: true
.For example, take the following configuration:
This configuration meets all of the requirements to share an nflog.
When an Alertmanager is run with this configuration, you can see from the
flushing alerts
andNotify success
lines that two aggregation groups are created and then flushed:However, reading the nflog file (after shutdown) shows just one nflog entry on disk:
However, when each route uses a different receiver then there are two entries on disk:
I expect this to create even more issues if the routes have different timers (i.e.
group_wait
,group_interval
andrepeat_interval
) or active and mute time intervals:What did you expect to see?
I expected to see a separate entry for each aggregation group. Here is the nflog from the first example, but instead running the code in this branch:
The text was updated successfully, but these errors were encountered: