[IMPALA-5537] Impala does not retry RPCs that fail in SSL_read()

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.9.0, Impala 2.10.0
• Fix Version/s: Impala 2.10.0
• Component/s: Backend
• Target Version: Impala 2.10.0
• Epic Color: ghx-label-5
• Assignee: Michael Ho
• Reporter: Lars Volker
• Created: 19/Jun/17 23:38
• Updated: 20/Feb/18 23:30
• Resolved: 21/Jun/17 18:29

Description

IMPALA-5388 changed the RPC retry logic to be much less aggressive. This increased the probability of failing queries under load when using SSL. We should consider retrying RPCs if the underlying socket throws an exception during SSL_read().

Issue Links

• is broken by: IMPALA-5388 wrong results under stress with secure cluster
• is related to: IMPALA-5576 Wrong Cancel() in QueryState::ReportExecStatusAux() can lead to coordinator hang
• relates to: IMPALA-6548 RPCs that fail with 'SSL_read: Connection timed out' should also be retried