After some testing, I've confirmed that this seems to behave
significantly better in terms of the number of failed requests due to
Tornado being the process of restarting compared with the previous
version, as each individual process is only down for a short time,
rather than all of them being down at once.
Zulip has had a small use of WebSockets (specifically, for the code
path of sending messages, via the webapp only) since ~2013. We
originally added this use of WebSockets in the hope that the latency
benefits of doing so would allow us to avoid implementing a markdown
local echo; they were not. Further, HTTP/2 may have eliminated the
latency difference we hoped to exploit by using WebSockets in any
case.
While we’d originally imagined using WebSockets for other endpoints,
there was never a good justification for moving more components to the
WebSockets system.
This WebSockets code path had a lot of downsides/complexity,
including:
* The messy hack involving constructing an emulated request object to
hook into doing Django requests.
* The `message_senders` queue processor system, which increases RAM
needs and must be provisioned independently from the rest of the
server).
* A duplicate check_send_receive_time Nagios test specific to
WebSockets.
* The requirement for users to have their firewalls/NATs allow
WebSocket connections, and a setting to disable them for networks
where WebSockets don’t work.
* Dependencies on the SockJS family of libraries, which has at times
been poorly maintained, and periodically throws random JavaScript
exceptions in our production environments without a deep enough
traceback to effectively investigate.
* A total of about 1600 lines of our code related to the feature.
* Increased load on the Tornado system, especially around a Zulip
server restart, and especially for large installations like
zulipchat.com, resulting in extra delay before messages can be sent
again.
As detailed in
https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it
appears that removing WebSockets moderately increases the time it
takes for the `send_message` API query to return from the server, but
does not significantly change the time between when a message is sent
and when it is received by clients. We don’t understand the reason
for that change (suggesting the possibility of a measurement error),
and even if it is a real change, we consider that potential small
latency regression to be acceptable.
If we later want WebSockets, we’ll likely want to just use Django
Channels.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Nowm unless you specify `--fill-cache`, memcached caches will not be
pre-filled after a server restart. This will be helpful when someone
is in a hurry (e.g. if the server is down right now, or if he/she
testing a configuration change in a newly setup server), it's best to
just restart without pre-filling the cache.
Fixes: #10900.
Previously, we unconditionally tried to restart the Tornado process
name corresponding to the historically always-true case of a single
Tornado process. This resulted in Tornado not being automatically
restarted on a production deployment on servers with more than one
Tornado process configured.
This dramatically reduces the Tornado downtime when restarting a Zulip
server, which is generally the most significant source of user-facing
bad experiences.
This is mostly important in that if you're running this as part of a
follow-up to a failed upgrade, and you don't do this,
process-fts-updates will be left not running, resulting in full-text
search not updating.
This commits adds the necessary puppet configuration and
installer/upgrade code for installing and managing the thumbor service
in production. This configuration is gated by the 'thumbor.pp'
manifest being enabled (which is not yet the default), and so this
commit should have no effect in a default Zulip production environment
(or in the long term, in any Zulip production server that isn't using
thumbor).
Credit for this effort is shared by @TigorC (who initiated the work on
this project), @joshland (who did a great deal of work on this and got
it working during PyCon 2017) and @adnrs96, who completed the work.
This causes `upgrade-zulip-from-git`, as well as a no-option run of
`tools/build-release-tarball`, to produce a Zulip install running
Python 3, rather than Python 2. In particular this means that the
virtualenv we create, in which all application code runs, is Python 3.
One shebang line, on `zulip-ec2-configure-interfaces`, explicitly
keeps Python 2, and at least one external ops script, `wal-e`, also
still runs on Python 2. See discussion on the respective previous
commits that made those explicit. There may also be some other
third-party scripts we use, outside of this source tree and running
outside our virtualenv, that still run on Python 2.
The manual step here is that we need to do the `puppet apply` before
pushing this commit, or `restart-server` will crash.
Previously we shut down everything in one group, which performed
poorly with supervisor's bad performance on restarting many daemons at
once. Now we shut down the unimportant stuff, then the important
stuff, bring back the important stuff, and then bring back the
unimportant stuff.
This new model has a little over 5s of downtime for the core
user-facing daemons -- which is still far more than would be ideal,
but a lot less than the 13s or so that we had before.
Here's some logs with the current setup for the tornado/django downtime:
2013-12-19 20:16:51,995 restart-server: Stopping daemons
2013-12-19 20:16:53,461 restart-server: Starting daemons
2013-12-19 20:16:57,146 restart-server: Starting workers
Compare with the behavior on master today:
2013-12-19 20:21:45,281 restart-server: Stopping daemons
2013-12-19 20:21:49,225 restart-server: Starting daemons
2013-12-19 20:21:58,463 restart-server: Done!
(imported from commit b2c1ba77f3dc989551d0939779208465a8410435)
This reverts commit acef4c0027b77053497ef6e9f7aa4b61703205c3.
Despite the lower total downtime, this caused more user-facing downtime.
(imported from commit 5cce032bb20abe83853a65ee72bf0bb28af403cc)
This shaves about 1.5 seconds off our restart time on ls-dev (9s ->
7.5s). Still too slow, but it's a little bit better.
(imported from commit acef4c0027b77053497ef6e9f7aa4b61703205c3)
Run the following commands as root before deploying this branch:
# /root/zulip/tools/migrate-server-config
# rm /etc/zulip/machinetype /etc/zulip/server /etc/zulip/local /etc/humbug-machinetype /etc/humbug-server /etc/humbug-local
(imported from commit aa7dcc50d2f4792ce33834f14761e76512fca252)
The moved files are:
humbug-server
humbug-local
humbug-machinetype
Their new names are their old names with 'humbug-' removed.
zulip-puppet-apply must be run before this commit is deployed
(imported from commit f4eb523244d3409b5809c279301225d3fdf0c230)
Anders Kaseorg
Tim Abbott
Anders Kaseorg
Harshit Bansal
Anders Kaseorg
Abhilash Verma
Joshua Schmidlkofer
rht
Greg Price
Umair Khan
Taranjeet Singh
Tim Abbott
Eklavya Sharma
Steven Oud
Tim Abbott
Zev Benjamin