[
https://issues.apache.org/jira/browse/PHOENIX-5250?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17284146#comment-17284146
]
Istvan Toth commented on PHOENIX-5250:
--------------------------------------
Looking at the changes, but without running any tests, I get the impression
that this cannot be fixed from the Phoenix side.
Fixing from the HBase side seems easy, if not free (again, not tested) by
changing the FSWALEntry constructor:
{code:java}
if (inMemstore) {
// construct familyNames here to reduce the work of log sinker.
Set<byte[]> families = edit.getFamilies();
// Phoenix uses METAFAMILY entries to persist index operations to the WAL,
// those must not be added to the memstore
families.remove(WALEdit.METAFAMILY);
this.familyNames = families != null ? families :
collectFamilies(edit.getCells());
} else {
this.familyNames = Collections.emptySet();
}
{code}
> The accumulated wal files can't be cleaned
> ------------------------------------------
>
> Key: PHOENIX-5250
> URL: https://issues.apache.org/jira/browse/PHOENIX-5250
> Project: Phoenix
> Issue Type: Bug
> Reporter: Jaanai Zhang
> Priority: Blocker
> Fix For: 5.1.0
>
> Attachments: image-2019-04-19-21-31-27-888.png
>
>
> Because of the modification of HBASE-20781, the faked WALEdits won't be
> filtered, all WALEdits will be saved into Memstore with a status that
> inMemstore is true(uses WAL->append method).
> !image-2019-04-19-21-31-27-888.png|width=755,height=310!
> The family array of IndexedKeyValue is WALEdit.METAFAMILY that is used to
> describe a fake WALEdit, and it will put into Memstore with WALedits of data
> table during sync global index.
> WAL files can't be cleaned, except for restarting RS, Many WAL files will
> decrease the percent of disk free.
> !https://gw.alicdn.com/tfscom/TB1n3cDQVzqK1RjSZFCXXbbxVXa.png|width=422,height=159!
>
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
