couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benoit Chesneau (JIRA)" <>
Subject [jira] [Commented] (COUCHDB-2097) Avoid performance regression with a single fd
Date Sat, 05 Jul 2014 15:28:34 GMT


Benoit Chesneau commented on COUCHDB-2097:

This is not about having 2 files process or not. With the change everything is now serialised,
first in the updater process and then in the file process.  So with this changes read and
writes are now waiting in the same queue. 2 times. This design is already a problem and the
one we should skip in a database generally. I already showed on a single node that long writes
(with large docs or bulk updates) have a direct impact on reads (i will work on a test case
to measure it). Latest rcouch have reverted this change and is quite more stable on a standalone
node.Imo this is a direct regression compared to the old situation where multiple reads can
be done distinctly from writes.

I know that this changes was done to remove the reference counting. However I don't see why
this should be done by only relying on 1 file process. Why not a database process? In any
case I think that removing or improving reference counting shouldn't have an impact on read//writes.
The current patch may improve the overall performances in the database but it adds a bottleneck
by doing it. 

I would be interested by measuring the difference on that particular part, not on the overall
performance of the db. So we can make a rational decisions pondering advantages of one against
the other and make sure most scenarios are taken in consideration.

> Avoid performance regression with a single fd
> ---------------------------------------------
>                 Key: COUCHDB-2097
>                 URL:
>             Project: CouchDB
>          Issue Type: Improvement
>      Security Level: public(Regular issues) 
>          Components: BigCouch
>            Reporter: Paul Joseph Davis
>              Labels: release
>         Attachments: mike-wallace-fd-benchmarks.txt
> Part of one of our large enhancements required that we remove a CouchDB performance optimization
on having two file handles to each .couch file. We need to make sure that this doesn't negatively
impact performance.

This message was sent by Atlassian JIRA

View raw message