[
https://issues.apache.org/jira/browse/MINIFICPP-1127?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Arpad Boda updated MINIFICPP-1127:
----------------------------------
Description:
There are several major issues in the current implementation of Provenance repo:
* Repo size is not maintained consistently, repo never gets full, new elements
are inserted without any restriction. Filling the underlying disc is just a
question of time.
* There is a logic to remove older records, but that's never triggered due to
improper size handling. (The counter never reaches the threshold limit)
* Even if it would be triggered it is very inefficient, resulting in cpu load
spikes.
* Write buffers are not configured properly, the result is sawtooth appearing
in memory consumption (up to 64MB).
was:
There are several major issues in the current implementation of Provenance repo:
Repo size is not maintained consistently, repo never gets full, new elements
are inserted without any restriction. Filling the underlying disc is just a
question of time.
There is a logic to remove older records, but that's never triggered due to
improper size handling. (The counter never reaches the threshold limit)
Even if it would be triggered it is very inefficient, resulting in cpu load
spikes.
Write buffers are not configured properly, the result is sawtooth appearing in
memory consumption (up to 64MB).
> Provenance repo performance should be improved
> ----------------------------------------------
>
> Key: MINIFICPP-1127
> URL: https://issues.apache.org/jira/browse/MINIFICPP-1127
> Project: Apache NiFi MiNiFi C++
> Issue Type: Bug
> Affects Versions: 0.7.0
> Reporter: Arpad Boda
> Assignee: Arpad Boda
> Priority: Major
> Fix For: 0.8.0
>
>
> There are several major issues in the current implementation of Provenance
> repo:
> * Repo size is not maintained consistently, repo never gets full, new
> elements are inserted without any restriction. Filling the underlying disc is
> just a question of time.
> * There is a logic to remove older records, but that's never triggered due to
> improper size handling. (The counter never reaches the threshold limit)
> * Even if it would be triggered it is very inefficient, resulting in cpu load
> spikes.
> * Write buffers are not configured properly, the result is sawtooth appearing
> in memory consumption (up to 64MB).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
