On 09/01/2025 14:01, ma...@apache.org wrote:
This is an automated email from the ASF dual-hosted git repository.
markt pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tomcat.git
The following commit(s) were added to refs/heads/main by this push:
new 1f4175d65c Refactor so the buffered data is used directly rather than
copied
1f4175d65c is described below
commit 1f4175d65cd7bb44af7fa098e8160b084870865a
Author: Mark Thomas <ma...@apache.org>
AuthorDate: Thu Jan 9 14:01:18 2025 +0000
Refactor so the buffered data is used directly rather than copied
This also aligns with other uses of o.a.catalina.connector.InputBuffer
where a pre-populated buffer replaces InputBuffer#bb rather than the
data being copied into the existing InputBuffer#bb
This is in preparation for a fix for the high HTTP/2 memory usage
reported on the users list before the holidays.
As far as I can tell, InputBuffer#bb - at least the 8kB buffer created
when the InputBuffer is created - is now never used. #setByteBuffer() is
always used to replace the existing buffer rather than copy data to the
existing buffer.
Recycle could/should be setting InputBuffer#bb to null to avoid
retaining references to these (potentially large) buffers but that
creates issues as there are multiple code paths that access
InputBuffer#bb when it is expected to empty. If recycle sets it to null,
we get NPEs.
My current plan is to create InputBuffer with bb set to a zero length
ByteBuffer and have recycle (re)set it to a zero length ByteBuffer. That
avoids the NPEs, avoids retaining references unnecessarily but (so far)
doesn't break anything.
I have some more testing to do but hope to have something to commit by
the end of today.
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
