[ https://issues.apache.org/jira/browse/IMPALA-6078?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Tim Armstrong resolved IMPALA-6078. ----------------------------------- Resolution: Duplicate Fix Version/s: Impala 2.9.0 > IMPALA query fails if a scratch directory goes bad with a disk failure > ---------------------------------------------------------------------- > > Key: IMPALA-6078 > URL: https://issues.apache.org/jira/browse/IMPALA-6078 > Project: IMPALA > Issue Type: Improvement > Components: Backend > Affects Versions: Impala 2.9.0 > Reporter: kamal Chandrasekaran > Fix For: Impala 2.9.0 > > > Impala was configured with two scratch directories on 2 different disks. > We simulated disk failure on one of the Scratch directories and it became > inaccessible > In such scenarios, We get the following error when a query tries to spill to > the scratch directories and the query fails: > >>>>>>>>> > > select avg(t1.a1), max(t2.a2) from impala_disk_test.impala_tst1 as t1 full > > join impala_disk_test.impala_tst1 as t2 ON t1.a1=t2.a1; > Query: select avg(t1.a1), max(t2.a2) from impala_disk_test.impala_tst1 as t1 > full join impala_disk_test.impala_tst1 as t2 ON t1.a1=t2.a1 > Cancelling Query > Application Exception : Default TException. > t2.a1;t.impala_tst1 as t1 full join impala_disk_test.impala_tst1 as t2 ON > t1.a1=t > Query: select avg(t1.a1), max(t2.a2) from impala_disk_test.impala_tst1 as t1 > full join impala_disk_test.impala_tst1 as t2 ON t1.a1=t2.a1 > WARNINGS: > No usable temporary files: space could not be allocated on any temporary > device. > Create file > /srv/impala2/impala/impala-scratch/5b431fe3571b9bf1:ef5b367fb74fd1b9_597f261f-792e-4ade-8b32-f4198c945142 > failed with errno=5 description=Error(5): Input/output error > No usable temporary files: space could not be allocated on any temporary > device. > >>>>>>>>> > It does not look good from HA point of view to have an Impala daemon in the > cluster that will sometimes fail queries just because it has a disk failed a > few hour ago. > If a disk fails during a query there is really nothing you can do, This > request is to address query failures that started after the disk has failed. -- This message was sent by Atlassian JIRA (v6.4.14#64029)