hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matteo Bertozzi (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-16554) Procedure V2 - handle corruption of WAL trailer
Date Fri, 02 Sep 2016 22:09:20 GMT

     [ https://issues.apache.org/jira/browse/HBASE-16554?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Matteo Bertozzi updated HBASE-16554:
------------------------------------
    Description: 
If the last wal was closed cleanly, the global tracker will be the last wal tracker (no rebuild
needed)
if the last wal does not have a tracker (corrupted/master-killed). on load() we will rebuild
the global tracker.

To compute quickly which files should be deleted, we also want the tracker of each file.
if the wal was closed properly and has a tracker we are good, if not we need to rebuild the
tracker for that file.
each file tracker contains a bitmap about what is in the wal (the updated bitmap), which is
easy to compute just by reading each entry of the wal. and a bitmap that keeps track of the
"running procedures" up to that wal (the deleted bitmap). The delete bitmap requires a bit
of post read-all-wals work. and it will basically require to AND the deleted bitmap of 

  was:if trailer is corrupted, build tracker for each WAL when loading, then use these trackers
to build global tracker.


> Procedure V2 - handle corruption of WAL trailer
> -----------------------------------------------
>
>                 Key: HBASE-16554
>                 URL: https://issues.apache.org/jira/browse/HBASE-16554
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Appy
>            Assignee: Appy
>         Attachments: tracker-rebuild.patch
>
>
> If the last wal was closed cleanly, the global tracker will be the last wal tracker (no
rebuild needed)
> if the last wal does not have a tracker (corrupted/master-killed). on load() we will
rebuild the global tracker.
> To compute quickly which files should be deleted, we also want the tracker of each file.
> if the wal was closed properly and has a tracker we are good, if not we need to rebuild
the tracker for that file.
> each file tracker contains a bitmap about what is in the wal (the updated bitmap), which
is easy to compute just by reading each entry of the wal. and a bitmap that keeps track of
the "running procedures" up to that wal (the deleted bitmap). The delete bitmap requires a
bit of post read-all-wals work. and it will basically require to AND the deleted bitmap of




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message