At some point the PostgreSQL Docker image started creating the zulip
database for us, which caused our CREATE DATABASE to fail.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
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>
Our recent fixes to using the system's configured memcached settings
broke populate_db, because its hacky clear_database helper is called
with a hacked-up settings module.
We fix this by first moving this out-of-place code from models.py into
populate_db, and then saving the settings required to access memcached
so that we can use them in clear_database.
We also fix a mypy erorr in flush-memcached that matches the same
issue fixed in clear_database.
This simplifies the RDS installation process to avoid awkwardly
requiring running the installer twice, and also is significantly more
robust in handling issues around rerunning the installer.
Finally, the answer for whether dictionaries are missing is available
to Django for future use in warnings/etc. around full-text search not
being great with this configuration, should they be required.
We'll be soon documenting a production workflow that involves using
it, and that means it needs to live under scripts/ (since tools/ isn't
present in release tarballs).
`copytree` throws an error if the target already exists, and we don’t
really want to rerun the copy anyway.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Ultimately, this isn't an effective way to monitor this queue; we want
time-based monitoring, not count-based monitoring. Doing that
properly will likely involve modifying the queue processor to write
something about its status.
But until we add the monitoring we want, it makes sense to leave this
active with low limits.
This is needed on at least Debian 10, otherwise xmlsec fails to
install: `Could not find xmlsec1 config. Are libxmlsec1-dev and
pkg-config installed?`
Also remove libxmlsec1-openssl, which libxmlsec1-dev already depends.
(No changes are needed on RHEL, where libxml2-devel and xmlsec1-devel
already declare a requirement on /usr/bin/pkg-config.)
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
The output log from running clean_unused_caches was too verbose as
part of the `upgrade-zulip` overall output. While this output is
potentially helpful when running it directly for debugging, it's
certainly redundant for the main production use case.
So a new flag --no-print-headers is introduced. It suppresses the
header outputs for the subtools.
Fixes#13214.
This allows the system to get updates to the Groonga repository
signing key, so `apt update` doesn’t start failing when the key
changes (like it recently did).
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
debian-archive-keyring is a dependency of the essential package apt,
so it is present in every Debian system.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
virtualenv on Ubuntu 16.04, when creating a new environment, downloads
the current version of setuptools, then replaces its pkg_resources
with an old copy from
/usr/share/python-wheels/pkg_resources-0.0.0-py2.py3-none-any.whl.
This causes problems, a simple example of which is reproducible from
the ubuntu:16.04 Docker base image as follows:
apt-get update
apt-get -y install python3-virtualenv
python3 -m virtualenv -p python3 /ve
/ve/bin/pip install sockjs-tornado
/ve/bin/pip download sockjs-tornado
→ `AttributeError: '_NamespacePath' object has no attribute 'sort'`
More relevantly, it breaks pip-compile in the same way. To fix this,
we need to force setuptools to be reinstalled, even if we’re asking
for the same version.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
This should dramatically improve the queue processor's performance in
cases where there's a very high volume of requests on a given endpoint
by a given user, as described in the new docstring.
Until we test this more broadly in production, we won't know if this
is a full solution to the problem, but I think it's likely. We've
never seen the UserActivityInterval worker end up backlogged without a
total queue processor outage, and it should have a similar workload.
Fixes#13180.
To replace DISTRIB_FAMILY, there’s now an os_families function using
the standard ID and ID_LIKE information in /etc/os-release.
Fixes#13070; fixes#13071.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
We no longer use tsearch_extras, and the camo patch is irrelevant on
systemd systems (Xenial and newer). So we no longer need to
provide/install a PPA at all.
Closes#13027.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Now that we're implemented tsearch_extras in pure postgres, we no
longer need a custom extension. This should help us considerably, as
it means we no longer need to ship custom apt packages at all.
Fixes#467.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Anders Kaseorg
Tim Abbott
rht
Vishnu KS
ab1nash
Mateusz Mandera