Re: haproxy 2.0-dev5-a689c3d - A bogus STREAM [0x805547500] is spinning at 100000 calls per second and refuses to die, aborting now!
Hi Pieter, On Fri, Jun 07, 2019 at 11:10:46PM +0200, PiBa-NL wrote: > Hi Willy, > > Op 7-6-2019 om 9:03 schreef Willy Tarreau: > > Hi again Pieter, > > > > On Tue, Jun 04, 2019 at 04:59:06PM +0200, Willy Tarreau wrote: > > > Whatever the values, no single stream should be woken up 100k times per > > > second or it definitely indicates a bug (spinning loop that leads to > > > reports > > > of 100% CPU)! > > > > > > I'll see if I can get something out of this. > > So just for the record, this is expected to be fixed in dev6 (it's the > > major change there). I'm interested in your feedback on this one, of > > course! > > > > Willy > > The stream does not spin anymore with dev6 so that seems to work alright. > Thanks. Excellent, thank you very much for your test and feedback! Willy
Re: haproxy 2.0-dev5-a689c3d - A bogus STREAM [0x805547500] is spinning at 100000 calls per second and refuses to die, aborting now!
Hi Willy, Op 7-6-2019 om 9:03 schreef Willy Tarreau: Hi again Pieter, On Tue, Jun 04, 2019 at 04:59:06PM +0200, Willy Tarreau wrote: Whatever the values, no single stream should be woken up 100k times per second or it definitely indicates a bug (spinning loop that leads to reports of 100% CPU)! I'll see if I can get something out of this. So just for the record, this is expected to be fixed in dev6 (it's the major change there). I'm interested in your feedback on this one, of course! Willy The stream does not spin anymore with dev6 so that seems to work alright. Thanks. Regards, PiBa-NL (Pieter)
Re: haproxy 2.0-dev5-a689c3d - A bogus STREAM [0x805547500] is spinning at 100000 calls per second and refuses to die, aborting now!
Hi again Pieter, On Tue, Jun 04, 2019 at 04:59:06PM +0200, Willy Tarreau wrote: > Whatever the values, no single stream should be woken up 100k times per > second or it definitely indicates a bug (spinning loop that leads to reports > of 100% CPU)! > > I'll see if I can get something out of this. So just for the record, this is expected to be fixed in dev6 (it's the major change there). I'm interested in your feedback on this one, of course! Willy
Re: haproxy 2.0-dev5-a689c3d - A bogus STREAM [0x805547500] is spinning at 100000 calls per second and refuses to die, aborting now!
Hi Pieter, On Tue, Jun 04, 2019 at 12:39:31AM +0200, PiBa-NL wrote: > Hi List, > > Hereby a message produced by haproxy 2.0-dev5-a689c3d that requested me to > write this mail ;) > > *** h1 1.4 debug|[ALERT] 154/000112 (90353) : A bogus STREAM > [0x805547500] is spinning at 10 calls per second and refuses to die, > aborting now! Please report this error to developers [strm=0x805547500 > src=::1 fe=fe1 be=be1 dst=srv1 rqf=4884a860 rqa=0 rpf=80408008 rpa=0 > sif=EST,80008 sib=EST,200111 af=0x0,0 csf=0x80552f180,4000 ab=0x0,0 > csb=0x80552f240,40 > cof=0x805528000,80201300:H1(0x8055500a0)/SSL(0x80552f000)/tcpv6(59) > cob=0x805528380,282306:H1(0x80555)/RAW(0x0)/tcpv4(60) ] Argh! Bad! I faced one of these yesterday when hammering compression *before* the commit you used, and this commit fixed the problem. I'm going to manually iterate the function with the elements above to see if something prevents it from finishing. > Scenario that led here.. : > -running 10 simultaneous tests 'tls_health_checks.vtc' 150 times total > (multiple times in a row..). > -machine its running on is a FreeBSD11 VM with 16 vCPU's so should be able > to handle 10 running haproxy's at 1 time, i think, it does get a bit busy > though, could it be caused by this possible 'miss configuration' / > congestion.?. (just tried again with '-j 5', and it happened again.) > -modified the timeout values of test: > /reg-tests/checks/tls_health_checks.vtc trying to get 'stable' results each > time, as you can see in the output some tests fail with 'unexpected' output > as far as defined in the test.. the values used are in the logoutput as well > but ill put them here as well: ( client 220 / server 200 /connect 20 > and client 220 / server 120 / connect 20 ) , not sure if these values > make sense, but still should not be possible to cause the spinning message > right.?. Whatever the values, no single stream should be woken up 100k times per second or it definitely indicates a bug (spinning loop that leads to reports of 100% CPU)! I'll see if I can get something out of this. Thank you! willy
