[
https://issues.apache.org/jira/browse/TIKA-2575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16364161#comment-16364161
]
Nicholas DiPiazza edited comment on TIKA-2575 at 2/14/18 3:01 PM:
------------------------------------------------------------------
This sentence is the kicker
> However, this may not prevent cases where we're currently loading large
>chunks of the document into memory before parsing.
In our code we have a "output-to-input-bytes ratio" where we kill if > a
threshold, say 200% of the size is a reasonable value.
Yet the thread in this snapshot
!screenshot-1.png!
Note the dominating thread is using 500 MB! Thus the crash.
So to prevent embedded jvms from crashing:
* must enforce a max file size to parse i.e. 50 MB
* must enforce a maximum input-to-output size ratio i.e. *2.0* for 200%
* must enforce a max number of tika parser threads
* must make sure jvm -Xmx is no lower than *(num threads) ** *(max files size)
* (max input-to-output ratio) * 5*
was (Author: ndipiazza_gmail):
This sentence is the kicker
> However, this may not prevent cases where we're currently loading large
>chunks of the document into memory before parsing.
In our code we have a "output-to-input-bytes ratio" where we kill if > a
threshold, say 200% of the size is a reasonable value.
Yet the thread in this snapshot
!screenshot-1.png!
Note the dominating thread is using 500 MB! Thus the crash.
So to prevent embedded jvms from crashing:
* must enforce a max file size to parse i.e. 50 MB
* must enforce a maximum input-to-output size ratio i.e. *2.0* for 200%
* must enforce a max number of tika parser threads
* must make sure jvm -Xmx is no lower than *(max files size) * (max
input-to-output ratio) * 5*
> Provide a way to abort tika parses when tika input stream buffer grows passed
> a certain threshold
> -------------------------------------------------------------------------------------------------
>
> Key: TIKA-2575
> URL: https://issues.apache.org/jira/browse/TIKA-2575
> Project: Tika
> Issue Type: Bug
> Components: parser
> Reporter: Nicholas DiPiazza
> Priority: Major
> Attachments: screenshot-1.png
>
>
> Sometimes, for example, you use tika to parse an XLS file that isn't really
> that big, maybe 60 MB. and suddenly the JVM heap size taken is >800Mb which
> causes an OOM in my case.
> Can we make an "abort threshold" where the tika parse will halt if parse
> output bytes exceeds this value?
> Or it is possible for users to already do this themselves by watching the
> input stream as it grows somehow?
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
