phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kadir OZDEMIR (Jira)" <>
Subject [jira] [Updated] (PHOENIX-5928) Index rebuilds without replaying data table mutations
Date Tue, 16 Jun 2020 20:56:00 GMT


Kadir OZDEMIR updated PHOENIX-5928:
    Attachment: PHOENIX-5928.4.x.007.patch

> Index rebuilds without replaying data table mutations
> -----------------------------------------------------
>                 Key: PHOENIX-5928
>                 URL:
>             Project: Phoenix
>          Issue Type: Improvement
>    Affects Versions: 5.0.0, 4.14.3
>            Reporter: Kadir OZDEMIR
>            Assignee: Kadir OZDEMIR
>            Priority: Major
>         Attachments: PHOENIX-5928.4.x.001.patch, PHOENIX-5928.4.x.002.patch, PHOENIX-5928.4.x.003.patch,
PHOENIX-5928.4.x.004.patch, PHOENIX-5928.4.x.005.patch, PHOENIX-5928.4.x.006.patch, PHOENIX-5928.4.x.007.patch
>          Time Spent: 4.5h
>  Remaining Estimate: 0h
> Index rebuilds are done currently by reading data table mutations (in UngroupedAggregateRegionObserver)
and then replaying them on the data table (in IndexRegionObser) -- without actually applying
mutations on the data table -- in order to generate the corresponding index mutations and
apply them on index tables. IndexRegionObserver sets the operation status for these data table
mutations to NOWRITE so that they are ignored by HBase after generating index table mutations
from these data table mutations. Since we do not apply these mutations on the data table,
there is no need to send these mutations to the data table regions if index mutations are
generated and sent to index tables by UngroupedAggregateRegionObserver. By doing so, we eliminate
going through data table update path and its overhead (its interaction with flushes, row locking

This message was sent by Atlassian Jira

View raw message