Kapacitor alert delays

Nobody has claimed this yet.

Hi,

We have been "fighting" against the delay on some kapacitor alerts over the past months. So far, we fix or understood some of those (problems on our scripts, problems on the data injected on influx...)

Now we are stuck with some new incidents that make noice on the entire organization. The thing is that, checking the logs over thhose alerts and its groups, kapacitor is making the same job twice at the same time:
ts=2018-08-30T07:45:00.068+01:00 lvl=info msg="begin batch" service=kapacitor task_master=main task=node_http_errors_exceeded node=influxdb_out15 prefix= name=js.http.requests group="environment=pro,jiraprojectname=project,platform=platform,targetDomain=domain,tenant=tenant" tag_environment=pro tag_jiraprojectname=project tag_platform="platform" tag_targetDomain=domain tag_tenant=tenant time=2018-08-30T06:40:00Z
ts=2018-08-30T07:45:00.068+01:00 lvl=info msg="batch point" service=kapacitor task_master=main task=node_http_errors_exceeded node=influxdb_out15 prefix= name=js.http.requests group="environment=pro,jiraprojectname=project,platform=platform,targetDomain=domain,tenant=tenant" tag_environment=pro tag_jiraprojectname=project tag_platform="platform" tag_targetDomain=domain tag_tenant=tenant field_http_errors=4 time=2018-08-30T06:40:00Z
ts=2018-08-30T07:45:00.068+01:00 lvl=info msg="end batch" service=kapacitor task_master=main task=node_http_errors_exceeded node=influxdb_out15 prefix= name=js.http.requests group="environment=pro,jiraprojectname=project,platform=platform,targetDomain=domain,tenant=tenant" tag_platform="platform" tag_targetDomain=domain tag_tenant=tenant tag_environment=pro tag_jiraprojectname=project time=2018-08-30T06:40:00Z
ts=2018-08-30T07:45:00.069+01:00 lvl=info msg="begin batch" service=kapacitor task_master=main task=node_http_errors_exceeded node=influxdb_out15 prefix= name=js.http.requests group="environment=pro,jiraprojectname=project,platform=platform,targetDomain=domain,tenant=tenant" tag_jiraprojectname=project tag_platform="platform" tag_targetDomain=domain tag_tenant=tenant tag_environment=pro time=2018-08-29T07:05:00Z
ts=2018-08-30T07:45:00.069+01:00 lvl=info msg="end batch" service=kapacitor task_master=main task=node_http_errors_exceeded node=influxdb_out15 prefix= name=js.http.requests group="environment=pro,jiraprojectname=project,platform=platform,targetDomain=domain,tenant=tenant" tag_environment=pro tag_jiraprojectname=project tag_platform="platform" tag_targetDomain=domain tag_tenant=tenant time=2018-08-29T07:05:00Z

So, mainly is going 2 days in the past at the same time is going for this data accurate in time...

We headed to the grouping, but as is is unuque, no more same combination exists!

Any idea or experience on something like this?

The issue provides only Kapacitor log excerpts and no source files, tests, or entry points. Start by identifying why the same task processes data from two different timestamps concurrently, then reproduce the delayed alert behavior and define the expected batch-processing result.

Written by the indexing model from the issue text.