Enis Soztutar updated HBASE-18565:
    Attachment: hbase-18565_v1.patch

v1 patch. The problem is that in AsyncScanRetryingCaller, the chain generated 
from Future.then() invocations results in a case where we are executing an RPC 
(thus trying to write to the pipeline) while still being inside the 
{{hbase::ClientHandler::read}} call chain (look at the above stack traces). The 
solution is moving the work for the new Scan.Next() call from IO thread to CPU 
executor thread. The patch also contains another bug fix. 

> [C++] Fix deadlock in AsyncScanRetryingCaller and other RPCs
> ------------------------------------------------------------
>                 Key: HBASE-18565
>                 URL: https://issues.apache.org/jira/browse/HBASE-18565
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Enis Soztutar
>            Assignee: Enis Soztutar
>             Fix For: HBASE-14850
>         Attachments: hbase-18565_v1.patch
> When running the load-client test, sometimes we get a deadlock.

This message was sent by Atlassian JIRA

Reply via email to