Bug #3204
closedCentOS 7 issue with lighttpd 1.4.70 (or other platforms using glibc <= 2.23)
Description
Updated to 1.4.70 from 1.4.69 on two production servers. Both are CentOS 7.
Server A with hundreds of request per minute, server B not "so many".
Server A got lighttpd not responding (but alive) after some minutes. This is behind haproxy (lighttpd backend via unix sockets)
Server A got reverted to lighttpd 1.4.69 (compiled from sources again) and is running for 24 hours long now.
Server B got several hours running fine, until it got not responding too after 8 hours. Reverted lighttpd to 1.4.69. It's OK now.
Just to file a report, I'll try to diagnose / see something with test servers and post back, obviously I can't afford doing tests on production. HAProxy is OK, the not responding server seems to be lighttpd / PHP and just reverting to 1.4.69 fixed things up.
Updated by juangacovas 12 months ago
I'll try to post more insights, this is H2 from HAProxy (SSL offloading), to H2 lighttpd (with no SSL module). Will post config and relevant info if it helps
Updated by juangacovas 12 months ago
gstrauss wrote in #note-3:
Please see #3201 There is a patch on lighttpd git master which fixes the code for glibc before glibc 2.24. Does that address your issue?
Thank you, I'm trying now the latest git branch, just the same way I tried the released 1.4.70, compiled with gcc (GCC) 11.2.1 20220127 (Red Hat 11.2.1-9)
- lighttpd -v
lighttpd/1.4.71-devel-lighttpd-1.4.70-5-ge05a935
So far so good for now.
My ./configure is: CFLAGS="-O2" --enable-mmap=yes --disable-ipv6 --with-brotli --with-maxminddb
I'll let you know after some hours and more traffic.
Updated by juangacovas 12 months ago
Everything seems good with the patch until now. Tomorrow (Monday) I could confirm around midday for "real", not weekend traffic in my use case, but I would say is ok now.
Updated by gstrauss 12 months ago
- Subject changed from Had to revert to 1.4.69 under CentOS 7 to CentOS 7 issue with lighttpd 1.4.70 (or other platforms using glibc <= 2.23)
- Status changed from New to Duplicate
The title of the original bug report was very, very poorly worded. Please avoid filing such bug reports.
"Had to revert to 1.4.69 under CentOS 7" is a very poorly worded title for a bug report. That describes an action you took -- it does not describe a bug.
If you have no idea if something is a bug or do not have any details describing the bug, then instead of filing a bug report, please ask a question in the Forums (see tab near top of page). "Something is not working" is vague. Do not file as a bug. Instead, ask a question in the forums. "<specific thing> stopped working in lighttpd 1.4.70" is a better bug report, if details are included in the description of the bug report.
Since the patch in #3201 appears to have addressed your issue, the problem (stemming from behavior in the ANCIENT glibc running in your Centos 7 with new code in lighttpd 1.4.70) would have caused any lighttpd 1.4.70 child processes to fail, including: include_shell in the config, "bin-path" to start backends, or CGI programs. Your bug report here about "Server A got lighttpd not responding (but alive) after some minutes." suggests to me that you did very little troubleshooting besides noting that your application was not responding. Please do not file bug reports for lighttpd asking for support troubleshooting your application. This is an issue tracker for lighttpd, not for end-user support requests. Ask questions in the forums.
Also available in: Atom