cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alex Petrov (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-12374) Can't rebuild SASI index
Date Thu, 18 Aug 2016 21:20:20 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-12374?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15427175#comment-15427175
] 

Alex Petrov commented on CASSANDRA-12374:
-----------------------------------------

You're right, sleep would be not reliable there. I've changed test to just write garbage to
file and make sure that it gets overwritten plus index is query-able after rebuild. So in
fact the test without sleep turned out to be much more useful.

I've triggered CI, will check results tomorrow morning, but local runs were okay.

One question: currently unqueriable indexes / are skipped. For example, if there's just one
sstable, and it's index is corrupted, it's going to get skipped and query will yield no results
rather than an exception. That might be desired though, just checking.

> Can't rebuild SASI index
> ------------------------
>
>                 Key: CASSANDRA-12374
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12374
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Alex Petrov
>            Assignee: Alex Petrov
>
> There's been no real requirement for that so far. 
> As [~beobal] has pointed out, it's not a big issue, since that only could be needed when
index files are lost, data corruption on disk (hardware issue) has occurred or there was a
bug that'd require an index rebuild.
> During {{rebuild_index}} task, indexes are only "marked" as removed with {{SecondaryIndexManager::markIndexRemoved}}
and then {{buildIndexesBlocking}} is called. However, since SASI keeps track of SSTables for
the index, it's going to filter them out with {{.filter((sstable) -> !sasi.index.hasSSTable(sstable))}}
in {{SASIIndexBuildingSupport}}.
> If I understand the logic correctly, we have to "invalidate" (drop data) right before
we re-index them. This is also a blocker for [CASSANDRA-11990] since without it we can't have
an upgrade path.
> I have a patch ready in branch, but since it's a bug, it's better to have it released
earlier and for all branches affected.
> cc [~xedin]



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

Mime
View raw message