[ https://issues.apache.org/jira/browse/HAWQ-275?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Dong Li updated HAWQ-275: ------------------------- Description: When we kill a QE of segment, the segment will restart all of its child process, and not have QE anymore. But the master still believes the QEs are cached, and it will dispatch these non-exist QE to handle query. The most problem is that if we have 6 QEs before kill them, and we want to execute a simple sql which only need 2 QEs. Then it will check and error three times and only after that it order segment to start QEs. {code} intern=# insert into b values (2 ); ERROR: Query Executor Error in seg4 localhost:40000 pid=19024: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); ERROR: Query Executor Error in seg0 localhost:40000 pid=19020: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); ERROR: Query Executor Error in seg2 localhost:40000 pid=19022: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); INSERT 0 1 {code} was: When we kill a QE of segment, the segment will restart all of its child process, and not have QE anymore. But the master still believes the QEs are cached, and it will dispatch these no-exist QE to handle query. The most problem is that if we have 6 QEs before kill them, and we want to execute a simple sql which only need 2 QEs. Then it will check and error three times and only after that it order segment to start QEs. {code} intern=# insert into b values (2 ); ERROR: Query Executor Error in seg4 localhost:40000 pid=19024: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); ERROR: Query Executor Error in seg0 localhost:40000 pid=19020: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); ERROR: Query Executor Error in seg2 localhost:40000 pid=19022: server closed the connection unexpectedly DETAIL: This probably means the server terminated abnormally before or while processing the request. intern=# insert into b values (2 ); INSERT 0 1 {code} > After killing QE of segment, the QE pool is not updated when dispatch > --------------------------------------------------------------------- > > Key: HAWQ-275 > URL: https://issues.apache.org/jira/browse/HAWQ-275 > Project: Apache HAWQ > Issue Type: Bug > Components: Dispatcher > Reporter: Dong Li > Assignee: Lei Chang > > When we kill a QE of segment, the segment will restart all of its child > process, and not have QE anymore. But the master still believes the QEs are > cached, and it will dispatch these non-exist QE to handle query. > The most problem is that if we have 6 QEs before kill them, and we want to > execute a simple sql which only need 2 QEs. Then it will check and error > three times and only after that it order segment to start QEs. > {code} > intern=# insert into b values (2 ); > ERROR: Query Executor Error in seg4 localhost:40000 pid=19024: server closed > the connection unexpectedly > DETAIL: > This probably means the server terminated abnormally > before or while processing the request. > intern=# insert into b values (2 ); > ERROR: Query Executor Error in seg0 localhost:40000 pid=19020: server closed > the connection unexpectedly > DETAIL: > This probably means the server terminated abnormally > before or while processing the request. > intern=# insert into b values (2 ); > ERROR: Query Executor Error in seg2 localhost:40000 pid=19022: server closed > the connection unexpectedly > DETAIL: > This probably means the server terminated abnormally > before or while processing the request. > intern=# insert into b values (2 ); > INSERT 0 1 > {code} -- This message was sent by Atlassian JIRA (v6.3.4#6332)