phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "James Taylor (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-4053) Lock row exclusively when necessary for mutable secondary indexing
Date Mon, 31 Jul 2017 00:02:33 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-4053?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16106705#comment-16106705
] 

James Taylor commented on PHOENIX-4053:
---------------------------------------

Looks like HBASE-17210 isn't available in 1.3. I could potentially hack around it by putting
LockManager in the org.apache.hadoop.hbase.ipc package, adding some casts, and referencing
deadline directly like this:
{code}
            RpcServer.Call call = (RpcServer.Call)RpcServer.getCurrentCall();
            if (call != null && call.deadline < Long.MAX_VALUE) {
              int timeToDeadline = (int)(call.deadline - System.currentTimeMillis());
              if (timeToDeadline <= this.rowLockWaitDuration) {
                reachDeadlineFirst = true;
                timeout = timeToDeadline;
              }
            }
{code}
WDYT, [~apurtell]? Too ugly, wait for 1.4 release for Phoenix, or is this ok?

> Lock row exclusively when necessary for mutable secondary indexing
> ------------------------------------------------------------------
>
>                 Key: PHOENIX-4053
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4053
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: James Taylor
>            Assignee: James Taylor
>         Attachments: PHOENIX-4053_4.x-HBase-0.98_v2.patch, PHOENIX-4053_4.x-HBase-0.98_v3.patch,
PHOENIX-4053-4.x-HBase-0.98_v4.patch, PHOENIX-4053_v2.patch, PHOENIX-4053_v3.patch, PHOENIX-4053_v4.patch,
PHOENIX-4053_wip.patch
>
>
> From HBase 1.2 on, rows are not exclusively locked when the preBatchMutate call is made
(see HBASE-18474). The mutable secondary index (global and local) depend on this to get a
consistent snapshot of a row between the point when the current row value is looked up, and
when the new row is written, until the mvcc is advanced. Otherwise, a subsequent update to
a row may not see the current row state. Even with pre HBase 1.2 releases, the lock isn't
held long enough for us. We need to hold the locks from the start of the preBatchMutate (when
we read the data table to get the prior row values) until the mvcc is advanced (beginning
of postBatchMutateIndispensably).
> Given the above, it's best if Phoenix manages the row locking itself (mimicing the current
HBase mechanism).



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message