[
https://issues.apache.org/jira/browse/NIFI-2934?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15609191#comment-15609191
]
Joseph Gresock commented on NIFI-2934:
--------------------------------------
Ok, I finally found the cause of the "too many files" issue. I found a post
that recommended cat /proc/<nifi pid>/limits, and found that when I start NiFi
using *salt* (a remote executor), NiFi starts with 4096 max open files.
However, if I start NiFi directly from the VM, either using the service or the
script, it starts with the configured system limits (50000).
This appears to be a known issue with salt, so I don't think it's NiFi's
responsibility. Therefore, I believe the only issue is the one that's probably
related to NIFI-2925, which was simply exacerbated by the file limit thing.
> Archiver still not respecting
> nifi.content.repository.archive.max.usage.percentage
> ----------------------------------------------------------------------------------
>
> Key: NIFI-2934
> URL: https://issues.apache.org/jira/browse/NIFI-2934
> Project: Apache NiFi
> Issue Type: Bug
> Affects Versions: 0.7.0, 0.7.1
> Reporter: Joseph Gresock
> Attachments: Disk-Usage-Increasing.png, NiFi-80-percent-disk.png,
> Queued.png, System-Diagnostics.png, content_repository usage.png, lsof.txt
>
>
> This seems related to NIFI-1726: we've noticed that the content repository
> takes up increasingly more space over time, even beyond the configured max
> usage percentage (see images). After restarting the NiFi cluster we get an
> immediate drop in disk usage with lots of log statements indicating that
> expired content is being removed.
> Not sure if this is related, but we also often get "Too many open files"
> during this expiration process after NiFi restart, despite lsof indicating a
> count far lower than our configured nofile and fs-max.
> In the environment indicated by the pictures,
> nifi.content.repository.archive.max.usage.percentage = 50%. Note that the
> flow itself only has ~240GB queued across the entire cluster, but each
> content_repository directory has over 360GB on each worker. Also note the
> disk usage graph increasing above 50% on each worker, until we finally
> restart and then the usage drops below 50%.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
