cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (CASSANDRA-7066) Simplify (and unify) cleanup of compaction leftovers
Date Tue, 28 Jul 2015 17:23:06 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-7066?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14644685#comment-14644685
] 

Benedict edited comment on CASSANDRA-7066 at 7/28/15 5:23 PM:
--------------------------------------------------------------

Well, a bug in the implementation could screw us up either way, and I'm not sure one is more
robust to it than any other. But I'm certainly not terribly opposed to changing it again.
It shouldn't be a drastic change.

However if we're doing that, I'd rather we just went with a simple log file that represents
new and old in one. i.e., we write lines like:

{noformat}
add:sstable-3
remove:sstable-2
commit
{noformat}

commit is only written very last if we are removing the old ones and adding the new ones.
Otherwise we rollback.

This makes the changes pretty minimal, as behaviourally it's identical, it's just the on-disk
representation that changes. It also retains the benefit of not double-counting your data.
If we want to be _really_ secure, we can post-fix each line with a checksum for the entire
file (up to the point), and if any do not match we retain every file as a last-ditch fallback.
We can also log panics in that case, so the operator knows for sure something bad is happening
with their filesystem. (if only the last line does not match, and it is not "commit", we're
as safe as we can be to rollback - but perhaps in this case we just log less panic-stricken
warnings that they can consider deleting the duplicate files).


was (Author: benedict):
Well, a bug in the implementation could screw us up either way, and I'm not sure one is more
robust to it than any other. But I'm certainly not terribly opposed to changing it again.
It shouldn't be a drastic change.

However if we're doing that, I'd rather we just went with a simple log file that represents
new and old in one. i.e., we write lines like:

{{noformat}}
add:sstable-3
remove:sstable-2
commit
{{noformat}}

commit is only written very last if we are removing the old ones and adding the new ones.
Otherwise we rollback.

This makes the changes pretty minimal, as behaviourally it's identical, it's just the on-disk
representation that changes. It also retains the benefit of not double-counting your data.
If we want to be _really_ secure, we can post-fix each line with a checksum for the entire
file (up to the point), and if any do not match we retain every file as a last-ditch fallback.
We can also log panics in that case, so the operator knows for sure something bad is happening
with their filesystem. (if only the last line does not match, and it is not "commit", we're
as safe as we can be to rollback - but perhaps in this case we just log less panic-stricken
warnings that they can consider deleting the duplicate files).

> Simplify (and unify) cleanup of compaction leftovers
> ----------------------------------------------------
>
>                 Key: CASSANDRA-7066
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7066
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Benedict
>            Assignee: Stefania
>            Priority: Minor
>              Labels: benedict-to-commit, compaction
>             Fix For: 3.0 alpha 1
>
>         Attachments: 7066.txt
>
>
> Currently we manage a list of in-progress compactions in a system table, which we use
to cleanup incomplete compactions when we're done. The problem with this is that 1) it's a
bit clunky (and leaves us in positions where we can unnecessarily cleanup completed files,
or conversely not cleanup files that have been superceded); and 2) it's only used for a regular
compaction - no other compaction types are guarded in the same way, so can result in duplication
if we fail before deleting the replacements.
> I'd like to see each sstable store in its metadata its direct ancestors, and on startup
we simply delete any sstables that occur in the union of all ancestor sets. This way as soon
as we finish writing we're capable of cleaning up any leftovers, so we never get duplication.
It's also much easier to reason about.



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

Mime
View raw message