Bird 1.6.3 cycle-restarting with SegFault during load-testing

[Maria Matějka]

Hello!

Please check your filedescriptor limit by
  ulimit -n
and check how many files are open by
  lsof -p <BIRD's pid>

BIRD has no such limit, it is capable of handling any reasonable (fitting in memory) number of file descriptors.

If you wanted any other help, feel free to ask more. BIRD should have no hard limits on scalability, yet some parts of it may be unexpectedly slow in extreme conditions.

Maria

On June 30, 2019 6:31:53 PM GMT+02:00, David.Garay at os3.nl wrote:
>Dear all, good afternoon
>
>New user, excuses in advance if the question has been answered
>elsewhere 
>- despite not having found anything on the mailing list history.
>
>For my Master thesis, I am investigating the impact of configuration 
>changes resulting from IRRDB updates on route server infrastructure. 
>During my experiments I got a segfault from Bird, and wanted to ask a 
>few questions about it.
>
>The details of the problem are below. Please let me know if you miss 
>anything, and thank you in advance,
>
>David
>
>
>** Problem description:
>For load testing, I am having my Bird route server instance peer with 
>other 1100 peers (containers, basic bgp config).
>
>CPU utilization and Memory are normal (~5%/30% respectively) and I see 
>BGP session establishments go above 1000 sessions without problems, but
>
>at 1013 consistently Bird raises seg-fault and restarts.
>
>Output shows:
>
>2019-06-27 12:37:49 <ERR> BGP: Error on listening socket: Too many open
>
>files
>
>Which seem to be related to dynamic buffers running out due to BGP open
>
>sessions.
>
>Bird version: 1.6.3
>CPU details: Intel(R) Xeon(R) CPU E3-1220L V2 @ 2.30GHz, 4 cores).
>
>
>** Actions taken:
>- Resource dump available, also: repeated tests to verify the restart 
>always happen at 1013 bgp sessions.
>
>** Questions:
>- Is the issue I just described known?
>- Are there (known) hard-coded limits on Bird? What is the expected 
>behaviour under these conditions?
>- In general, what are other dimensioning variables to be taken into 
>account with Bird?
>
>** Logs:
>Zip file containing: resource dump, count of bgp sessions, further 
>analysis.
>https://we.tl/t-mRmgIgC38a

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://trubka.network.cz/pipermail/bird-users/attachments/20190701/bdb3854d/attachment.htm>