Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
ghx-label-14
Description
Looking through the code for ImpalaServer, it looks like there are several threads / thread-pools that are started on executor-only impalads, but don't actually do anything:
- The "cancellation-worker" threadpool (ImpalaServer::CancelFromThreadPool)
- The "session-maintenance" thread (ImpalaServer::SessionMaintenance)
- The "query-expirer" thread (ImpalaServer::ExpireQueries)
- The "unresponsive-backend-thread" thread (ImpalaServer::UnresponsiveBackendThread)
- The code to start ImpalaInternalService might be dead code, but maybe we should just delete it since the ImpalaInternalService doesn't exist anymore
Confirmed this my creating a cluster with dedicated coordinator and getting a thread dump of an executor, which showed the following:
Thread 16 (Thread 0x7fe96fe57700 (LWP 8721)): #0 0x00007fea1ad20360 in pthread_cond_wait@@GLIBC_2.3.2 () from /lib/x86_64-linux-gnu/libpthread.so.0 No symbol table info available. #1 0x0000000001ce4423 in impala::ConditionVariable::Wait (this=0x110b3370, lock=...) at /home/stakiar/Impala/be/src/util/condition-variable.h:49 mutex = 0x110b3348 #2 0x0000000002469f6a in impala::ImpalaServer::SessionMaintenance (this=0x110b3200) at /home/stakiar/Impala/be/src/service/impala-server.cc:2055 timeout_lock = {_M_device = 0x110b3348, _M_owns = true} now = 140640581412928 expired_cnt = 0
Attachments
Issue Links
- is related to
-
IMPALA-9609 Minimize Frontend activity in executor only Impalas
- Resolved