cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-8707) Move SegmentedFile, IndexSummary and BloomFilter to utilising RefCounted
Date Fri, 06 Feb 2015 13:13:35 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-8707?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14309098#comment-14309098
] 

Benedict commented on CASSANDRA-8707:
-------------------------------------

bq. Also, I think we need some high level documentation with concrete examples about the life
cycle of sstables. ... (perhaps as a big comment on top in SSTableReader?)

This seems like an excellent idea, and this ticket seems like a good opportunity to introduce
it, although it will have to be revised with each of my follow ons. It didn't really occur
to me to do this since that is all unchanged here, but it has grown overtime to be very non-obvious.
It would be helpful if, once I've introdiced a high-level overview, the individuals familiar
with the _tool_ or special compaction lifecycles (or others) could amend the SSTableReader
description to make it clear how they are expected to behave and treat sstables, and what
assumptions they utilise, as that is often subtly different and I am not at all an expert
in those areas.

bq. I think the penny has finally dropped for me, the thing I'm missing now is more "why?"-comments,
for example, why do we need to copy CompressionMetadata in CM.Writer#open(..) if finish type
is final and not otherwise?

Could you list the places you think need this service? Since the code outside of SSTableReader
tidying is largely functionally unchanged, only swapping in a new API, it's not clear which
bits need this, outside of the high level overview. This particular spot, for instance, had
this behaviour previously (and it's about resizing, rather than copying, since we now know
the final size of the data so can release it). One of the follow tickets also cleans this
particular spot up a little, by introducing an enum specific to it that describes the reasons
for opening (which should make that more obvious). I'm hoping a small follow up I've yet to
publish will also simplify that particular bit further. I'm more than happy to flesh out comments,
but would appreciate an outline of which bits you think have been lacking historically.

> Move SegmentedFile, IndexSummary and BloomFilter to utilising RefCounted
> ------------------------------------------------------------------------
>
>                 Key: CASSANDRA-8707
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8707
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Benedict
>            Assignee: Benedict
>            Priority: Critical
>             Fix For: 2.1.3
>
>
> There are still a few bugs with resource management, especially around SSTableReader
cleanup, esp. when intermixing with compaction. This migration should help. We can simultaneously
"simplify" the logic in SSTableReader to not track the replacement chain, only to take a new
reference to each of the underlying resources.



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

Mime
View raw message