[ 
https://issues.apache.org/jira/browse/NUTCH-2929?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17473091#comment-17473091
 ] 

Sebastian Nagel commented on NUTCH-2929:
----------------------------------------

It wasn't that many Tika warnings:
- (non-parsing Fetcher, 160 threads) 400-500 during the first minute, then up 
to 50 for the next 2-4 minutes, nothing then
- (parsing Fetcher, 80 threads) ~2000 during the first minute, disappearing 
entirely within few minutes.

Note: Tika is used for MIME type detection whether Fetcher is parsing or not-

> Fetcher: start threads slowly to avoid that resources are temporarily 
> exhausted
> -------------------------------------------------------------------------------
>
>                 Key: NUTCH-2929
>                 URL: https://issues.apache.org/jira/browse/NUTCH-2929
>             Project: Nutch
>          Issue Type: Improvement
>    Affects Versions: 1.18
>            Reporter: Sebastian Nagel
>            Priority: Minor
>             Fix For: 1.19
>
>
> Fetcher spins all threads without any delay. This may cause that certain 
> resources are temporarily exhausted if all threads start fetching the first 
> pages simultaneously.
> The issue has been observed by Tika warnings about overuse of the SAXParser 
> pool which appeared only during the first 2-5 minutes of fetching a segment. 
> See https://lists.apache.org/thread/lo6b9wdlxy2lz12wmosldgl9x9ov1cks - adding 
> a short delay between thread launches makes the warnings disappear.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

Reply via email to