couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joan Touzet <woh...@apache.org>
Subject Re: [PROPOSAL] Officially deprecate CouchDB 1.x.
Date Fri, 06 Jul 2018 18:01:23 GMT
Hi Johs,

----- Original Message -----
> the flaw in measuring 2.x adoption by Github issues and requests for
> informal and paid support is that all these data points are pain
> points,
> not signs of error-free operation.
> The in-production part of CouchDB based systems is not easily
> measured, as one of the qualities of CouchDB has been zero downtime.
> The lack of data should just add caution before using the kill
> switch.

Point taken, though there are still plenty of problems and bugs in 1.x,
and we're not seeing people come and ask for help with them in Slack, IRC,
on these lists or privately.

Perhaps a better approach is to see the number of downloads of CouchDB
tarballs and packages, yes? We have access to this through our bintray.com
hosting and the apache.org mirror hierarchy.

For the former, see graphs here: https://imgur.com/a/nI2bvnx

Graph 1 shows downloads of the Windows package, which has the longest
download history available. Based on this we can see that the very large
majority of downloads of CouchDB is 2.x since the 2.0.0 release, and that
downloads had a palpable increase with this new version. 2017 still had ~10% of
downloads as 1.x releases, but by the time 2.1.0 released last July,
this became background noise.

With the release of 2.1.1 (see graph 2), 1.x accounts for less than 5%
of the downloads.

Graphs 3 and 4 are the same data for the Mac downloads. Total downloads
are only 20% of the Windows downloads, but the overall trend of data is
the same. In fact, on OSX, in 2018 downloads of 2.1.0 alone exceed
downloads of 1.6.1, even after 2.1.1-1 was published.

While I was at it, I checked download number for the new .deb/.rpm
packages (though we only provide 2.x packages). In the year the packages
have been available, we have ~20k downloads each on rpm and deb, with
debian users upgrading much more aggressively to 2.1.1 from 2.0.0 than
rpm users. See graphs 5 and 6.

There is the possibility that Windows downloads are skewed, since I
expect this is more frequently in use on desktops rather than servers.
With 1.x packages removed from many distributions due to the age of
SpiderMonkey 1.8.5, our best bet is to look at tarball downloads from
the Apache mirror sites. Apache Infra logs requests through the web that
redirect people to their nearest mirror; this is the best data we have.
The exposed file:

https://www-us.apache.org/dyn/stats/couchdb.log

only covers the last 30 days; I've asked Infra if they can provide logs
farther back for a better statistical sample.

Based on this file, using this program:

https://gist.github.com/wohali/899a6d6c316d2f52c7a9cec0b3b41580

the count of downloads is:

0.8: 7
0.9: 18
0.11: 6
0.10: 6
1.0: 34
1.1: 41
1.2: 50
1.3: 19
1.4: 20
1.5: 23
1.6: 127
1.7: 201
2.0: 28
2.1: 717

As expected, there is a higher 1.x count here, but it still accounts for
less than half of the downloads, even when taking into account archived
versions < 1.7.

If we only consider 1.6, 1.7, 2.0 and 2.1, 1.x downloads account for 
30% of tarball downloads.

These download numbers are further dwarfed by the binary downloads above.


Turning to those older distributions, installations of CouchDB 1.x on
Debian have dropped significantly in the last couple of years:

https://qa.debian.org/popcon.php?package=couchdb

Ubuntu's data is skewed, due to the inclusion of CouchDB 1.1(1.2?) in
very old versions of Ubuntu One that is no longer used. Even accounting
for that, the data is similar, showing less than 500 active users of
the package:

#Format
#   
#<name> is the package name;
#<inst> is the number of people who installed this package;
#<vote> is the number of people who use this package regularly;
#<old> is the number of people who installed, but don't use this package
#        regularly;
#<recent> is the number of people who upgraded this package recently;
#<no-files> is the number of people whose entry didn't contain enough
#        information (atime and ctime were 0).
#rank name                            inst  vote   old recent no-files (maintainer)
1209  couchdb-bin                    835764   453 834984    18   309 (Unknown) 

Both distributions have removed CouchDB from their repos in the latest
release.

The very old Ubuntu Launchpad CouchDB PPA with 1.x packages has an API for
querying the information, but I've not looked into it, because apparently
it can take hours to gather the stats and I'm unconvinced the data will
vary significantly from the easier-to-consume paths shown above.

https://ftagada.wordpress.com/2011/01/05/ppa-stats-initial-impressions/


Conclusion: Based on this information there is poor evidence for the
hypothesis that CouchDB 1.x is being downloaded with any sort of frequency.
Of course, this does not mean that people don't already have it downloaded
and are continuing to use it on already provisioned computing resources.


> Yes, I am volunteering for a hypothetical 3-man team. I understand
> that
> such a team would not pull resources allready committed to 2.x, so
> the
> hope would be that presently non-actice developers could be persuaded
> to join an effort with limited, but a forward-looking scope. Noone
> should
> be expected to work for a branch that is doomed to die.
> If the PMC support for and effort to keep 1.x alive can be expressed
> clearly, I can develop a discussion paper for a hypotetical team as a
> start.

Thanks for the offer. Before writing up a full paper, what would your
first 3 acts be?

-Joan

Mime
View raw message