[jira] [Updated] (PDFBOX-4668) Add ResourceCacheFactory as global setting

["Andreas Lehmkühler (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/PDFBOX-4668?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Andreas Lehmkühler updated PDFBOX-4668:
---------------------------------------
    Affects Version/s: 3.0.4 PDFBox
                       4.0.0

> Add ResourceCacheFactory as global setting
> ------------------------------------------
>
>                 Key: PDFBOX-4668
>                 URL: https://issues.apache.org/jira/browse/PDFBOX-4668
>             Project: PDFBox
>          Issue Type: Task
>          Components: Rendering
>    Affects Versions: 3.0.4 PDFBox, 4.0.0
>            Reporter: Ben Manes
>            Assignee: Andreas Lehmkühler
>            Priority: Major
>         Attachments: Screenshot 2023-03-20 at 18.57.40.png, memory.png, threads.png
>
>
> Image rendering is cached by {{DefaultResourceCache}} per-document using soft references. As described in the [FAQ|https://pdfbox.apache.org/2.0/faq.html], this can lead to an {{OutOfMemoryError}} when processing, e.g. many documents in parallel. The configuration of this cache is per-document and it is initialized with the default.
> {code}
> // document-wide cached resources
> private ResourceCache resourceCache = new DefaultResourceCache();
> {code}
> This requires all call sites be modified to disable it, some of which may be in 3rd party code. The ask is to static factory to configure the default globally, which would return a new {{DefaultResourceCache}} when called. This would let a user specify a new static factory, e.g. one that returns a custom cache or {{null}} if disabled.
> Soft references are a problematic caching scheme, which degrades poorly. It is very likely that the many and large image fragments causes GC promotion (eden=>young=>old) which requires a full GC to collect. Under memory/cpu pressure, the GC can devolve into a death spiral of collecting the minimal heap space to match its pause time constraints, leading to repeated GCs due to soft reference pollutions and an eventual OOME. If caching was set, it might be preferable to be size-based (by rough byte-size) and perhaps tied into {{MemoryUsageSetting}} main memory configuration.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@pdfbox.apache.org
For additional commands, e-mail: dev-help@pdfbox.apache.org