Обсуждение: PQConsumeinput stuck on recv
Hi, I've been experiencing an issue. We use an open-source VoIP software whose backend is PostgreSQL. Initially we had a twin-server setup (one server running the VoIP software, another one running the pg instance). Due to company growth we were running into performance issues, so we rolled out a new architecture using multiple VoIP servers connected to the single pg instance. Since then, the VoIP software started misbehaving - it randomly stops responding, and only a restart gets it back up running. It is random throughout the servers, time-of-day, day-of-week... we haven't found a correlation between it and any other metric like CPU usage, network traffic and such.
Andre
Since it's been happening for a few weeks now, every time it freezes we take a gcore dump and check it in gdb... and after a lot of hair pulling and learning about the innards of the VoIP software we see that most often the software is stuck in this call trace:
#0 in __libc_recv (fd=409, buf=0x7f2c4802e6c0, n=16384, flags=1898970523) at ../sysdeps/unix/sysv/linux/x86_64/recv.c:33
#1 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5
#2 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5
#3 in PQconsumeInput () from /usr/lib/x86_64-linux-gnu/libpq.so.5
The software shares a database connection between threads, and controls its access through a mutex, so once one thread that acquires the mutex gets stuck in the location above, all other threads starts pilling up behind the mutex, and that's apparently the reason the software stops responding for most of its functions (while other functions that do not depend on the database works normally).
And it stays stuck on it forever... at one situation we took two gcore dumps spaced 10 minutes apart, and they look almost identical, with the same thread stuck on recv and all the others waiting for the lock to be released.
I wonder if anyone has any tip on what to look for next... Besides the implementation of the VoIP software itself, we are looking into network issues (we are seeing a bunch of TCP retransmissions between some servers and the db), but otherwise no other app running on those servers has presented any weird behavior like this VoIP software. We don't understand what would cause recv to get stuck like this.
BTW we're running debian 9, pg 9.6.3, and the VoIP sofware (along most of the other apps) uses libpq of a slightly older version (9.4.15).
Thanks in advance.
Hi, On 2018-02-23 13:33:18 -0300, Andre Oliveira Freitas wrote: > Since it's been happening for a few weeks now, every time it freezes we > take a gcore dump and check it in gdb... and after a lot of hair pulling > and learning about the innards of the VoIP software we see that most often > the software is stuck in this call trace: > > #0 in __libc_recv (fd=409, buf=0x7f2c4802e6c0, n=16384, flags=1898970523) > at ../sysdeps/unix/sysv/linux/x86_64/recv.c:33 > #1 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5 > #2 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5 > #3 in PQconsumeInput () from /usr/lib/x86_64-linux-gnu/libpq.so.5 So it's just receiving data from the network. Have you verified whether the connection is actually stable? Any chance it's just waiting for the network to time out. Might be worth configuring tcp timeouts, to make sure its unrelated to that. What is the server showing as activity while the client is waiting? Could you show the corresponding pg_stat_activity row? > The software shares a database connection between threads, and controls its > access through a mutex, so once one thread that acquires the mutex gets > stuck in the location above, all other threads starts pilling up behind the > mutex, and that's apparently the reason the software stops responding for > most of its functions (while other functions that do not depend on the > database works normally). Hm, have you compiled libpq with threading support? Or use a distribution that compiles it with that? While I don't see an obvious connection to that stacktrace it seems worthwhile to verify. A mutex protecting individual connection, while also necessary if connections are shared, doesn't achieve the same. > I wonder if anyone has any tip on what to look for next... Any chance you're occasionally forking and then interacting with the connection in the forked process? Greetings, Andres Freund
2018-02-23 14:20 GMT-03:00 Andres Freund <andres@anarazel.de>: > > Hi, > > On 2018-02-23 13:33:18 -0300, Andre Oliveira Freitas wrote: > > Since it's been happening for a few weeks now, every time it freezes we > > take a gcore dump and check it in gdb... and after a lot of hair pulling > > and learning about the innards of the VoIP software we see that most often > > the software is stuck in this call trace: > > > > #0 in __libc_recv (fd=409, buf=0x7f2c4802e6c0, n=16384, flags=1898970523) > > at ../sysdeps/unix/sysv/linux/x86_64/recv.c:33 > > #1 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5 > > #2 in ?? () from /usr/lib/x86_64-linux-gnu/libpq.so.5 > > #3 in PQconsumeInput () from /usr/lib/x86_64-linux-gnu/libpq.so.5 > > So it's just receiving data from the network. Have you verified whether > the connection is actually stable? Any chance it's just waiting for the > network to time out. Might be worth configuring tcp timeouts, to make > sure its unrelated to that. > > What is the server showing as activity while the client is waiting? > Could you show the corresponding pg_stat_activity row? Sure, datid;datname;pid;usesysid;usename;application_name;client_addr;client_hostname;client_port;backend_start;xact_start;query_start;state_change;wait_event_type;wait_event;state;backend_xid;backend_xmin;query 16,4;freeswitch;13,322;16,386;freeswitch;freeswitch;192.168.0.101;;43,276;2018-02-20 14:11:56;;2018-02-20 14:24:15;2018-02-20 14:24:15;;;idle;;;COMMIT 16,4;freeswitch;13,323;16,386;freeswitch;freeswitch;192.168.0.101;;43,328;2018-02-20 14:11:57;;2018-02-20 14:24:57;2018-02-20 14:24:57;;;idle;;;COMMIT 16,4;freeswitch;13,416;16,386;freeswitch;freeswitch;192.168.0.101;;51,681;2018-02-20 14:16:02;;2018-02-20 14:16:31;2018-02-20 14:16:31;;;idle;;;insert into sip_authentication (nonce,expires,profile_name,hostname, last_nc) values('363d02f6-cb9a-4791-9e05-d18473a18812', 1519147649, 'internal', 'srv-1', 0) 16,4;freeswitch;13,472;16,386;freeswitch;freeswitch;192.168.0.101;;33,481;2018-02-20 14:22:13;;2018-02-20 14:25:09;2018-02-20 14:25:09;;;idle;;;select command from aliases where alias='show status' The problematic connection is the third one; by checking the timings on query_start it appears to be the last query executed by the stuck connection before it became stuck. There are no outstanding locks in any of the tables the VoIP software normally uses. > > > > The software shares a database connection between threads, and controls its > > access through a mutex, so once one thread that acquires the mutex gets > > stuck in the location above, all other threads starts pilling up behind the > > mutex, and that's apparently the reason the software stops responding for > > most of its functions (while other functions that do not depend on the > > database works normally). > > Hm, have you compiled libpq with threading support? Or use a > distribution that compiles it with that? While I don't see an obvious > connection to that stacktrace it seems worthwhile to verify. > > A mutex protecting individual connection, while also necessary if > connections are shared, doesn't achieve the same. I'm using the libpq that comes with debian, however I can install the library from the official repository to be sure, I assume the one from the official repo has it enabled. > > > I wonder if anyone has any tip on what to look for next... > > Any chance you're occasionally forking and then interacting with the > connection in the forked process? As far as I know, no. The software forks at the beginning but from then on, its just threads. > > Greetings, > > Andres Freund If it is of any help, here is the link to the implementation that consumes PQconsumeInput: https://freeswitch.org/stash/projects/FS/repos/freeswitch/browse/src/switch_pgsql.c#128 It is a function to check if the database connection is up before running a query. I'm not a mantainer nor an expert in pg, but we reviewed the implementation and seems OK. One thing that is bothering me by looking at the gdb backtraces is that recv always seems to be receiving a non-zero value in flags, even though libpq seems to pass zero. I don't know if its of any relevance. -- Thanks in advance, Andre
Hi, On 2018-02-23 15:59:18 -0300, Andre Oliveira Freitas wrote: > I'm using the libpq that comes with debian, however I can install the > library from the official repository to be sure, I assume the one from > the official repo has it enabled. That should be fine. Could you get the backtrace again, after installing debugging symbols? It'd certainly be helpful to see the exact path to the blocking syscall. Greetings, Andres Freund
Hi, I was able to capture the backtrace again, now with libpq debugging symbols. Thread 15 (Thread 0x7f8cec068700 (LWP 68)): #0 0x00007f8d252a1d9b in __libc_recv (fd=150, buf=0x7f8cf0034410, n=16384, flags=623517083, flags@entry=0) at ../sysdeps/unix/sysv/linux/x86_64/recv.c:33 #1 0x00007f8d26689783 in recv (__flags=0, __n=<optimized out>, __buf=<optimized out>, __fd=<optimized out>) at /usr/include/x86_64-linux-gnu/bits/socket2.h:44 #2 pqsecure_raw_read (conn=conn@entry=0x7f8cf001e390, ptr=<optimized out>, len=<optimized out>) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:230 #3 0x00007f8d26689863 in pqsecure_read (conn=conn@entry=0x7f8cf001e390, ptr=<optimized out>, len=<optimized out>) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:217 #4 0x00007f8d266810ea in pqReadData (conn=conn@entry=0x7f8cf001e390) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-misc.c:683 #5 0x00007f8d2667e6f2 in PQconsumeInput (conn=0x7f8cf001e390) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-exec.c:1704 In this case, I also checked the pg_stat_activity and this particular connection on pg server side was idle for 15 minutes. I killed it using pg_terminate_backend, and then somehow the application resumed execution. As you can see, recv has received a non-zero value in flags, which has the MSG_WAITALL bit is set, however in this case the application expects the read to be non-blocking. Thanks in advance, André Freitas 2018-02-23 16:10 GMT-03:00 Andres Freund <andres@anarazel.de>: > Hi, > > On 2018-02-23 15:59:18 -0300, Andre Oliveira Freitas wrote: >> I'm using the libpq that comes with debian, however I can install the >> library from the official repository to be sure, I assume the one from >> the official repo has it enabled. > > That should be fine. Could you get the backtrace again, after installing > debugging symbols? It'd certainly be helpful to see the exact path to > the blocking syscall. > > Greetings, > > Andres Freund -- André Luis O. Freitas System Architect Rua do Rócio, 220 - Cj. 72 São Paulo - SP - 04552-000 55 11 4063 4222 afreitas@callix.com.br www.callix.com.br
Hi, On 2018-03-01 19:20:04 -0300, Andre Oliveira Freitas wrote: > I was able to capture the backtrace again, now with libpq debugging symbols. > > Thread 15 (Thread 0x7f8cec068700 (LWP 68)): > #0 0x00007f8d252a1d9b in __libc_recv (fd=150, buf=0x7f8cf0034410, > n=16384, flags=623517083, flags@entry=0) at > ../sysdeps/unix/sysv/linux/x86_64/recv.c:33 > #1 0x00007f8d26689783 in recv (__flags=0, __n=<optimized out>, > __buf=<optimized out>, __fd=<optimized out>) at > /usr/include/x86_64-linux-gnu/bits/socket2.h:44 > #2 pqsecure_raw_read (conn=conn@entry=0x7f8cf001e390, ptr=<optimized > out>, len=<optimized out>) at > /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:230 > #3 0x00007f8d26689863 in pqsecure_read > (conn=conn@entry=0x7f8cf001e390, ptr=<optimized out>, len=<optimized > out>) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:217 > #4 0x00007f8d266810ea in pqReadData (conn=conn@entry=0x7f8cf001e390) > at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-misc.c:683 > #5 0x00007f8d2667e6f2 in PQconsumeInput (conn=0x7f8cf001e390) at > /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-exec.c:1704 > > In this case, I also checked the pg_stat_activity and this particular > connection on pg server side was idle for 15 minutes. I killed it > using pg_terminate_backend, and then somehow the application resumed > execution. There's something decidedly weird going on. Libpq always keeps the connection in nonblocking mode internally. Blocking is implemented by using select on the socket. So this should never block. Is there any chance parts of your application changes the sockets block-y-ness? Is see your code is passing the socket around, so perhaps that's happening somewhere outside of the file? > As you can see, recv has received a non-zero value in flags Well, no, not really. recv() has a 0 flags, it's just libc's internal implementation that appears to be showing up weird afaict. Greetings, Andres Freund
Ok, on one hand glad to hear that is an artifact, on the other hand that means the issue is still out there... I'm not a maintainer of Freeswitch, I am an user of it. However I am trying to supply the maintainers with any relevant information regarding this issue. As you mentioned, I see the socket being modified and passed around in the source code. Do you think the implementation is OK? I do not know if that's standard practice when consuming libpq. 2018-03-02 17:36 GMT-03:00 Andres Freund <andres@anarazel.de>: > Hi, > > > On 2018-03-01 19:20:04 -0300, Andre Oliveira Freitas wrote: >> I was able to capture the backtrace again, now with libpq debugging symbols. >> >> Thread 15 (Thread 0x7f8cec068700 (LWP 68)): >> #0 0x00007f8d252a1d9b in __libc_recv (fd=150, buf=0x7f8cf0034410, >> n=16384, flags=623517083, flags@entry=0) at >> ../sysdeps/unix/sysv/linux/x86_64/recv.c:33 >> #1 0x00007f8d26689783 in recv (__flags=0, __n=<optimized out>, >> __buf=<optimized out>, __fd=<optimized out>) at >> /usr/include/x86_64-linux-gnu/bits/socket2.h:44 >> #2 pqsecure_raw_read (conn=conn@entry=0x7f8cf001e390, ptr=<optimized >> out>, len=<optimized out>) at >> /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:230 >> #3 0x00007f8d26689863 in pqsecure_read >> (conn=conn@entry=0x7f8cf001e390, ptr=<optimized out>, len=<optimized >> out>) at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-secure.c:217 >> #4 0x00007f8d266810ea in pqReadData (conn=conn@entry=0x7f8cf001e390) >> at /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-misc.c:683 >> #5 0x00007f8d2667e6f2 in PQconsumeInput (conn=0x7f8cf001e390) at >> /build/postgresql-10-rIfDLC/postgresql-10-10.2/build/../src/interfaces/libpq/fe-exec.c:1704 >> >> In this case, I also checked the pg_stat_activity and this particular >> connection on pg server side was idle for 15 minutes. I killed it >> using pg_terminate_backend, and then somehow the application resumed >> execution. > > There's something decidedly weird going on. Libpq always keeps the > connection in nonblocking mode internally. Blocking is implemented by > using select on the socket. So this should never block. > > Is there any chance parts of your application changes the sockets > block-y-ness? Is see your code is passing the socket around, so perhaps > that's happening somewhere outside of the file? > > >> As you can see, recv has received a non-zero value in flags > > Well, no, not really. recv() has a 0 flags, it's just libc's internal > implementation that appears to be showing up weird afaict. > > > Greetings, > > Andres Freund -- André Luis O. Freitas System Architect Rua do Rócio, 220 - Cj. 72 São Paulo - SP - 04552-000 55 11 4063 4222 afreitas@callix.com.br www.callix.com.br
Hi, On 2018-03-05 17:57:51 -0300, Andre Oliveira Freitas wrote: > Ok, on one hand glad to hear that is an artifact, on the other hand > that means the issue is still out there... > > I'm not a maintainer of Freeswitch, I am an user of it. However I am > trying to supply the maintainers with any relevant information > regarding this issue. > > As you mentioned, I see the socket being modified and passed around in > the source code. Do you think the implementation is OK? I do not know > if that's standard practice when consuming libpq. It's ok to pass the socket around, it'd not be ok to change whether the socket blocks or not. I don't have the time to review the freeswitch code to see whether it does so. - Andres