[ 
https://issues.apache.org/jira/browse/KUDU-2031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16041547#comment-16041547
 ] 

Alexey Serbin commented on KUDU-2031:
-------------------------------------

Some print-outs of the scanned rows when the expected behavior was not observed 
(run via dist-test with --stress_cpu_threads=8 the binaries built from 
https://gerrit.cloudera.org/#/c/7037/10 in DEBUG):

READ_AT_SNAPSHOT with CLOSEST_REPLICA:
{noformat}
Value of: row_strings.size()
  Actual: 5
Expected: kNumTabletServers * (i + 1)
Which is: 6
(int32 key=0, int32 val=0) (int32 key=2, int32 val=2) (int32 key=1, int32 
val=1) (int32 key=5, int32 val=2) (int32 key=3, int32 val=0)
{noformat}
The expected set of keys: {{0, 1, 2, 3, 4, 5}}, the actual: {{0, 2, 1, 5, 3}}; 
i.e. key 4 is missing.

After adding the fault-tolerant property for scanners (i.e. calling 
{{KuduScanner::SetFaultTolerant()}}):

READ_AT_SNAPSHOT with CLOSEST_REPLICA, the fault-tolerant property is set:
{noformat}
Value of: row_strings.size()
  Actual: 4
Expected: kNumTabletServers * (i + 1)
Which is: 6
(int32 key=0, int32 val=0) (int32 key=2, int32 val=2) (int32 key=1, int32 
val=1) (int32 key=3, int32 val=0)
{noformat}
The expected set of keys: {{0, 1, 2, 3, 4, 5}}, the actual: {{0, 2, 1, 3}}; 
i.e. keys 4 and 5 are missing.

> READ_AT_SNAPSHOT scans with {CLOSEST_REPLICA, FIRST_REPLICA} selectors do not 
> yield RYW behavior in some scenarios
> ------------------------------------------------------------------------------------------------------------------
>
>                 Key: KUDU-2031
>                 URL: https://issues.apache.org/jira/browse/KUDU-2031
>             Project: Kudu
>          Issue Type: Bug
>          Components: client, tserver
>    Affects Versions: 1.4.0
>            Reporter: Alexey Serbin
>              Labels: consistency
>
> Running the updated scenario 
> {{ClientFailoverOnNegotiationTimeoutITest.Kudu1580ConnectToTServer}} from 
> {{client-negotiation-failover-itest}} sometimes fails to read the freshly 
> written rows if running READ_AT_SNAPSHOT scan with {{CLOSEST_REPLICA, 
> FIRST_REPLICA}} selection modes.
> To reproduce, run patch from https://gerrit.cloudera.org/#/c/7037/8 built in 
> DEBUG configuration.  The patch might be applied to the recent sources from 
> the main trunk as of 2017-06-03 (e.g. 
> e3b7d4dc1a6be020f3f126a66192ca1451e95e6f).



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Reply via email to