[
https://issues.apache.org/jira/browse/COUCHDB-568?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12776593#action_12776593
]
Paul Joseph Davis commented on COUCHDB-568:
-------------------------------------------
Definitely an interesting idea. In the thread on parallelized b~trees I was
basically thinking that we could take Damien's modifications to
couch_db_updater.erl and push all that logic down into couch_btree.erl which
would allow multiple mappers to make more efficient use of btree updates in
view generation.
Allowing a b~tree to hold new nodes in RAM for some duration before being
synced should allow for similar speedups in the condensed tree writes like
batch docs did.
Though, this could allow for readers to see a database in a state that was
never on disk. With batch writes we never allowed readers to see the docs in
the write buffer, so the progression of db state was unidirectional. With
pending writes viewable, an error could make the btree state go backwards. If
that makes sense.
Much to contemplate.
> When delayed_commits = true, keep updated btree nodes in memory until the
> commit
> --------------------------------------------------------------------------------
>
> Key: COUCHDB-568
> URL: https://issues.apache.org/jira/browse/COUCHDB-568
> Project: CouchDB
> Issue Type: Improvement
> Affects Versions: 0.10
> Reporter: Adam Kocoloski
>
> rnewson reported on IRC that the new batch=ok implementation results in
> significantly larger overhead in the .couch files. This makes sense; the old
> batch mode waited 1 second before saving, but the new implementation just
> updates the doc asynchronously. With fast hardware and moderate write rates
> it's likely that each document is being written separately.
> The overhead presumably arises from frequently updated btree inner nodes
> being written to disk many times over. I'm interested in exploring a
> modification of the delayed_commits mode whereby the updated btree nodes are
> not actually written to disk immediately, but are instead held in memory
> until the commit. I'd like to think that this will result in more compact
> files without any decrease in durability. New read requests would still be
> able to access these in-memory nodes.
> I realize the notion that updates go directly to disk is baked pretty deeply
> into couch_btree, but I still thought this was worth bringing up to a wider
> audience.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
