Description of problem:
Current implementation of the semaphore_wait() function is not fully thread-safe
Version-Release number of selected component (if applicable):
sometimes, when server is under heavy load
Steps to Reproduce:
1. send many queries for RRs which authoritative zones are served via bind-dyndb-ldap plugin
2. wait for server lockup
all server's worker threads are blocked in semaphore_wait()
Created attachment 516506 [details]
Created attachment 517022 [details]
pstack of hung named
I recently upgraded DNS my server from an older dual core core CPU to a quad core Xeon E3 and now named locks up after the first few queries. Stack trace attached.
I was able to work around the problem by reducing the number of worker threads named starts by adding the following line to /etc/sysconfig/named:
(In reply to comment #2)
> Created attachment 517022 [details]
> pstack of hung named
> I recently upgraded DNS my server from an older dual core core CPU to a quad
> core Xeon E3 and now named locks up after the first few queries. Stack trace
> I was able to work around the problem by reducing the number of worker threads
> named starts by adding the following line to /etc/sysconfig/named:
> OPTIONS="-n 1"
Which version of bind, bind-libs and bind-dyndb-ldap do you use, please?
I was experiencing the same semaphore error as described in the freeipa-users list. Other than serving records for IPA, my bind install is also an authoritative DNS server.
I rebuilt bind + bind-dyndb-ldap from Adam's proposed patches:
The resolver has now been running for 6+ hours, where previously it would hang on the semaphore issue within an hour.
Given that all of the servers we have in QA have not been locking up, even under load, and that this patch was submitted quite a while ago, I am going to mark this bug as verified.
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.