Broker connect getaddrinfo crash with concurrent producer and consumer

[voelkerp]

Description

I have made an executable that creates one rd_kafka_t as producer and one as consumer in separate threads. When using only one at a time, they work correctly. When I create both at the same time, the executable crashes on the second call to rd_kafka_brokers_add(). I'm only adding a single broker but I do it once for each rd_kafka_t.

I've tried setting the broker via "bootstrap.servers" and that has the same effect.

The crash occurs under rd_kafka_broker_thread_main() and getaddrinfo() in internal_getent() (see entire callstack below.)

I'm unsure what else to try. Any help or tips appreciated.

Checklist

• librdkafka version (release number or git tag): 0.11.5
• librdkafka client configuration: internal.termination.signal=SIGIO
• Operating system: Ubuntu 18.04.1 LTS

0x00007ffff0131448 in internal_getent (stream=stream@entry=0x7ffff002e670, result=result@entry=0x7fffd1ff8900, buffer=buffer@entry=0x7fffd1ff8d10 "",
buflen=buflen@entry=1024, errnop=errnop@entry=0x7fffd1ffb6b0, herrnop=herrnop@entry=0x7fffd1ffb6e8, af=0, flags=0) at nss_files/files-XXX.c:216
216 nss_files/files-XXX.c: No such file or directory.
(gdb) i stack
#0 0x00007ffff0131448 in internal_getent (stream=stream@entry=0x7ffff002e670, result=result@entry=0x7fffd1ff8900, buffer=buffer@entry=0x7fffd1ff8d10 "",
buflen=buflen@entry=1024, errnop=errnop@entry=0x7fffd1ffb6b0, herrnop=herrnop@entry=0x7fffd1ffb6e8, af=0, flags=0) at nss_files/files-XXX.c:216
#1 0x00007ffff01328bd in _nss_files_gethostbyname4_r (name=0x7fffd1ffa6a0 "xnode.second-bridge.net", pat=0x7fffd1ff8a88, buffer=0x7fffd1ff8d10 "", buflen=1024,
errnop=0x7fffd1ffb6b0, herrnop=0x7fffd1ffb6e8, ttlp=0x0) at nss_files/files-hosts.c:416
#2 0x0000000000791be5 in gaih_inet.constprop ()
#3 0x0000000000793b74 in getaddrinfo ()
#4 0x00000000004492ae in rd_getaddrinfo (nodesvc=, defsvc=0x7fffd1ffa7a0 "9092", flags=32, family=, socktype=,
protocol=, errstr=0x7fffd1ff9288) at ../../../Shared/librdkafka/src/rdaddr.c:168
#5 0x000000000046014a in rd_kafka_broker_resolve (rkb=) at ../../../Shared/librdkafka/src/rdkafka_broker.c:667
#6 rd_kafka_broker_connect (rkb=) at ../../../Shared/librdkafka/src/rdkafka_broker.c:1466
#7 rd_kafka_broker_thread_main (arg=0x7fffd8002dd0) at ../../../Shared/librdkafka/src/rdkafka_broker.c:3512
#8 0x00000000004be976 in _thrd_wrapper_function (aArg=) at ../../../Shared/librdkafka/src/tinycthread.c:624
#9 0x00000000004e0ebb in start_thread (arg=0x7fffd1ffb700) at pthread_create.c:463
#10 0x0000000000798f5f in clone ()

[rolandyoung]

Our application uses both consumer and producer instances simultaneously, but we create them one at a time from the same thread - the librdkafka code uses its own threads for all the hard work. Maybe serialising the startup would work around any thread safety issue in getaddrinfo.

[voelkerp]

Turns out I was compiling a static executable with "-static". There's documented issues with getaddrinfo and static compilation.

It works fine now that I've removed that compile switch. This issue can be closed.