return inside finally blocks causes exceptions to be silenced.
Although these blocks follow blanket ‘except Exception’ handlers, they
do not seem to have a goal of silencing BaseException and exceptions
thrown by the exception handler, so rewrite them to avoid it.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Both of these compatibility blocks can be deleted, since you can't
upgrade directly to any supported release from the versions where the
old event formats would be used.
Previously, an active production Zulip server would experience a class
of deadlocks caused by two or more concurrent bulk update operations
on the UserMessage table.
This is because UPDATE ... SET ... WHERE statements that execute in
parallel take row-level UPDATE locks as they get results; since the
query plans may result in getting rows in different orders between two
queries, this can result in deadlocks.
Some databases allow ORDER BY on their UPDATE ... WHERE statements;
PostgreSQL does not. In PostgreSQL, the answer is to do a sub-select
with an ORDER BY ... FOR UPDATE to ensure consistent ordering on row
locks.
We do this all code paths using bitand or bitor as part of bulk
editing message flags, which should ensure that these concurrent
operations obtain row level locks on the table in the same order.
Fixes#19054.
cfcbf58cd1 rightly removed the use of `user_ids` in
`render_markdown`, which in turn makes it unnecessary in
`render_incoming_message`.
Remove the unnecessary parameter from `render_incoming_message`.
We leave the fetching of links outside of the lock, as they could take
seconds, which is an unreasonable amount of time to hold a lock on the
message row. This may result in unnecessary work, in the case that
the message was since edited, but the unnecessary work is preferable
to blocking other work on the message row for the duration.
This commit changes the invite API to accept invitation
expiration time in minutes since we are going to add a
custom option in further commits which would allow a user
to set expiration time in minutes, hours and weeks as well.
The `get_link_embed_data` / `link_embed_data_from_cache` pair as
introduced in c93f1d4eda uses the cache
as a temporary store inside of the `embed_links` worker; this means
that it must be durable storage, or the worker will stall and re-fetch
the same links to preview them.
Switch to plumbing through the fetched URL embed data as an parameter
to the Markdown evaluation which uses them, rather than using the
cache as an intermediary. This frees up the cache to be merely a
non-durable cache.
As a side-effect, this removes get_cache_with_key, and
link_embed_data_from_cache which was its only callsite.
The database value for expiry_date is None for the invite
that will never expire and the clients send -1 as value
in the API similar to the message retention setting.
Also, when passing invite_expire_in_days as an argument
in various functions, invite_expire_in_days is passed as
-1 for "Never expires" option since invite_expire_in_days
is an optional argument in some functions and thus we cannot
pass "None" value.
As a preparatory step to refactoring json_success to accept
request as a parameter, change `do_report_error`, which is
called from the events queue for "error_reports", to return
None instead of json_success.
Adds an assertion error to `ErrorReporter` queue processor
and removes `JsonableError` from `do_report_error`.
It is likely that `do_error_report` was moved from a view in a
previous refactor, but was not updated to no longer return an
HttpReponse.
Python's behaviour on `sys.exit` is to wait for all non-daemon threads
to exit. In the context of the missedmessage_emails worker, if any
work is pending, a non-daemon Timer thread exists, which is waiting
for 5 seconds. As soon as that thread is serviced, it sets up another
5-second Timer, a process which repeats until all
ScheduledMessageNotificationEmail records have been handled. This
likely takes two minutes, but may theoretically take up to a week
until the thread exits, and thus sys.exit can complete.
Supervisor only gives the process 30 seconds to shut down, so
something else must prevent this endless Timer.
When `stop` is called, take the lock so we can mutate the timer.
However, since `stop` may have been called from a signal handler, our
thread may _already_ have the lock. As Python provides no way to know
if our thread is the one which has the lock, make the lock a
re-entrant one, allowing us to always try to take it.
With the lock in hand, cancel any outstanding timers. A race exists
where the timer may not be able to be canceled because it has
finished, maybe_send_batched_emails has been called, and is itself
blocked on the lock. Handle this case by timing out the thread join
in `stop()`, and signal the running thread to exit by unsetting the
timer event, which will be detected once it claims the lock.
RabbitMQ clients have a setting called prefetch[1], which controls how
many un-acknowledged events the server forwards to the local queue in
the client. The default is 0; this means that when clients first
connect, the server must send them every message in the queue.
This itself may cause unbounded memory usage in the client, but also
has other detrimental effects. While the client is attempting to
process the head of the queue, it may be unable to read from the TCP
socket at the rate that the server is sending to it -- filling the TCP
buffers, and causing the server's writes to block. If the server
blocks for more than 30 seconds, it times out the send, and closes the
connection with:
```
closing AMQP connection <0.30902.126> (127.0.0.1:53870 -> 127.0.0.1:5672):
{writer,send_failed,{error,timeout}}
```
This is https://github.com/pika/pika/issues/753#issuecomment-318119222.
Set a prefetch limit of 100 messages, or the batch size, to better
handle queues which start with large numbers of outstanding events.
Setting prefetch=1 causes significant performance degradation in the
no-op queue worker, to 30% of the prefetch=0 performance. Setting
prefetch=100 achieves 90% of the prefetch=0 performance, and higher
values offer only minor gains above that. For batch workers, their
performance is not notably degraded by prefetch equal to their batch
size, and they cannot function on smaller prefetches than their batch
size.
We also set a 100-count prefetch on Tornado workers, as they are
potentially susceptible to the same effect.
[1] https://www.rabbitmq.com/confirms.html#channel-qos-prefetch
The `current_queue_size` key in the queue monitoring stats file was
the local queue size, not the global queue size -- d5a6b0f99a
renamed the function, but did not adjust the queue monitoring JSON,
despite the last use of it having been removed in cd9b194d88.
The function is still used to mark "we emptied our queue," and it
remains a reasonable metric for that.
For export realm following changes have been made:
- `./manage.py export --upload` would delete `.tar.gz` and unpacked dir
- `./manage.py export` would only delete `unpacked dir`
Besides, we have removed `--delete-after-upload` as we have set it as
the default.
Fixes#20081
Since 3853285241, PushNotificationsWorker uses the aioapns library
to send Apple push notifications. This introduces an asyncio event
loop into this worker process, which, if unlucky, can respond poorly
when a SIGALRM is introduced to it:
```
[asyncio] Task exception was never retrieved
future: <Task finished coro=<send_apple_push_notification.<locals>.attempt_send() done, defined at /path/to/zerver/lib/push_notifications.py:166> exception=WorkerTimeoutException(30, 1)>
Traceback (most recent call last):
File "/path/to/zerver/lib/push_notifications.py", line 169, in attempt_send
result = await apns_context.apns.send_notification(request)
File "/path/to/zulip-py3-venv/lib/python3.6/site-packages/aioapns/client.py", line 57, in send_notification
response = await self.pool.send_notification(request)
File "/path/to/zulip-py3-venv/lib/python3.6/site-packages/aioapns/connection.py", line 407, in send_notification
response = await connection.send_notification(request)
File "/path/to/zulip-py3-venv/lib/python3.6/site-packages/aioapns/connection.py", line 189, in send_notification
data = json.dumps(request.message, ensure_ascii=False).encode()
File "/usr/lib/python3.6/json/__init__.py", line 238, in dumps
**kw).encode(obj)
File "/usr/lib/python3.6/json/encoder.py", line 199, in encode
chunks = self.iterencode(o, _one_shot=True)
File "/usr/lib/python3.6/json/encoder.py", line 257, in iterencode
return _iterencode(o, 0)
File "/path/to/zerver/worker/queue_processors.py", line 353, in timer_expired
raise WorkerTimeoutException(limit, len(events))
zerver.worker.queue_processors.WorkerTimeoutException: Timed out after 30 seconds processing 1 events
```
...which subsequently leads to the worker failing to make any progress
on the queue.
Remove the timeout on the worker. This may result in failing to make
forward progress if Apple/Google take overly long handling requests,
but is likely preferable to failing to make forward progress if _one_
request takes too long and gets unlucky with when the signal comes
through.
This extends the invite api endpoints to handle an extra
argument, expiration duration, which states the number of
days before the invitation link expires.
For prereg users, expiration info is attached to event
object to pass it to invite queue processor in order to
create and send confirmation link.
In case of multiuse invites, confirmation links are
created directly inside do_create_multiuse_invite_link(),
For filtering valid user invites, expiration info stored in
Confirmation object is used, which is accessed by a prereg
user using reverse generic relations.
Fixes#16359.
This information can be gleaned from the stacktrace, but making it
explicit in the stringification makes it much easier to differentiate
types of errors at a glance, particularly in Sentry.
maybe_send_batched_emails handles batches of emails from different
users at once; as it processes each user's batch, it enqueues messages
onto the `email_senders` queue. If `handle_missedmessage_emails`
raises an exception when processing a single user's email, no events
are marked as handled -- including those that were already handled and
enqueued onto `email_senders`. This results in an increasing number
of users being sent repeated emails about the same missed messages.
Catch and log any exceptions when handling an individual user's
events. This guarantees forward progress, and that notifications are
sent at-most-once, not at-least-once.
We only have one query which will change database state in this function,
and we already have a lock on the process itself, so there's no need for
a transaction.
This was added in ebb4eab0f9.
Previously, we stored up to 2 minutes worth of email events in memory
before processing them. So, if the server were to go down we would lose
those events.
To fix this, we store the events in the database.
This is a prep change for allowing users to set custom grace period for
email notifications, since the bug noted above will aggravate with
longer grace periods.
The `# nocoverage` was unnecessary apart from for the compatibility code,
so add a test for that code and remove the `# nocoverage`.
The `message_id` -> `message_ids` conversion was done in
9869153ae8.
This fixes a bug introduced in 95b46549e1
which made the worker simply log a warning about the timeout and then
continue consume()ing the event that should have also been interrupted.
The idea here is to introduce an exception which can be used to
interrupt the consume() process without triggering the regular handling
of exceptions that happens in _handle_consume_exception.
Throwing an exception is excessive in case of this worker, as it's
expected for it to time out sometimes if the urls take too long to
process.
With a test added by tabbott.
This allows specific queue workers to override the defaut behavior and
implement their own response to the timer expiring. We will want to use
this for embed_links queue at least.
This adds a new class called MessageRenderingResult to contain the
additional properties we added to the Message object (like alert_words)
as well as the rendered content to ensure typesafe reference. No
behavioral change is made except changes in typing.
This is a preparatory change for adding django-stubs to the backend.
Related: #18777
This is a prep commit for adding realm-level default for various
user settings. We add the language, in which the invite email will
be sent, to the dict added to queue itself to avoid making queries
in a loop when sending multiple emails from queue.
We also handle the case for old events in the queue.
Django's default SMTP implementation can raise various exceptions
when trying to send an email. In order to allow Zulip calling code
to catch fewer exceptions to handle any cause of "email not
sent", we translate most of them into EmailNotDeliveredException.
The non-translated exceptions concern the connection with the
SMTP server. They were not merged with the rest to keep some
details about the nature of these.
Tests are implemented in the test_send_email.py module.
This will stop dropping events in the case that the background
`maybe_send_batched_email` thread takes longer than 30s. However, see
also #15280 and the TODO comment about how we lose events upon
restart; this worker is still lossy.
Previously the outgoing emails were sent over several SMTP
connections through the EmailSendingWorker; establishing a new
connection each time adds notable overhead.
Redefine EmailSendingWorker worker to be a LoopQueueProcessingWorker,
which allows it to handle batches of events. At the same time, persist
the connection across email sending, if possible.
The connection is initialized in the constructor of the worker
in order to keep the same connection throughout the whole process.
The concrete implementation of the consume_batch function is simply
processing each email one at a time until they have all been sent.
In order to reuse the previously implemented decorator to retry
sending failures a new method that meets the decorator's required
arguments is declared inside the EmailSendingWorker class. This
allows to retry the sending process of a particular email inside
the batch if the caught exception leaves this process retriable.
A second retry mechanism is used inside the initialize_connection
function to redo the opening of the connection until it works or
until three attempts failed. For this purpose the backoff module
has been added to the dependencies and a test has been added to
ensure that this retry mechanism works well.
The connection is closed when the stop method is called.
Fixes: #17672.
This was introduced in 8321bd3f92 to serve as a sort of drop-in
replacement for zerver.lib.queue.queue_json_publish, but its use has
been subsequently cut out (e.g. `9fcdb6c83ac5`).
Remote its last callsite.
django.utils.translation.ugettext is a deprecated alias of
django.utils.translation.gettext as of Django 3.0, and will be removed
in Django 4.0.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
The existing organization, of returning an opaque blob from
`build_bot_request`, which was later consumed by
`send_data_to_server`, is not particularly sensible; the steps become
oddly split between the OutgoingWebhookWorker, `do_rest_call`, and the
`OutgoingWebhookServiceInterface`.
Make the `OutgoingWebhookServiceInterface` in charge of building,
making, and returning the request in one method; another method
handles extracting content from a successful response. `do_rest_call`
is responsible for calling both halves of this, and doing common error
handling.
The `deployment` key was only set in `do_report_error`, which is now
only used in one codepath (the queue worker). The logging handlers on
staging call notify_server_error directly, which omits the
`deployment` key.
Remove the odd one-of key, and instead simply do dispatch in
`do_report_error`.
Not all of the workers are known to be safe to interrupt; they might
leave inconsistent state. As such, terminating them with timeouts
should currently only be a last-resort against stalled queues, not a
regular occurrence.
Since the exception can be triggered at arbitrary places in the stack
based on whenever the alarm happens to fire, they do not often group
together.
Explicitly group them together, grouped only by which queue the work
is in.
We already trust ids that are put on our queue
for deferred work. For example, see the code for
"mark_stream_messages_as_read_for_everyone"
We now pass stream_recipient_id when we queue
up work for do_mark_stream_messages_as_read.
This generally saves about 3 queries per
user when we unsubscribe them from a stream.
Since this was using repead individual get() calls previously, it
could not be monitored for having a consumer. Add it in, by marking
it of queue type "consumer" (the default), and adding Nagios lines for
it.
Also adjust missedmessage_emails to be monitored; it stopped using
LoopQueueProcessingWorker in 5cec566cb9, but was never added back
into the set of monitored consumers.
This low-level interface allows consuming from a queue with timeouts.
This can be used to either consume in batches (with an upper timeout),
or one-at-a-time. This is notably more performant than calling
`.get()` repeatedly (what json_drain_queue does under the hood), which
is "*highly discouraged* as it is *very inefficient*"[1].
Before this change:
```
$ ./manage.py queue_rate --count 10000 --batch
Purging queue...
Enqueue rate: 11158 / sec
Dequeue rate: 3075 / sec
```
After:
```
$ ./manage.py queue_rate --count 10000 --batch
Purging queue...
Enqueue rate: 11511 / sec
Dequeue rate: 19938 / sec
```
[1] https://www.rabbitmq.com/consumers.html#fetching
Despite its name, the `queue_size` method does not return the number
of items in the queue; it returns the number of items that the local
consumer has delivered but unprocessed. These are often, but not
always, the same.
RabbitMQ's queues maintain the queue of unacknowledged messages; when
a consumer connects, it sends to the consumer some number of messages
to handle, known as the "prefetch." This is a performance
optimization, to ensure the consumer code does not need to wait for a
network round-trip before having new data to consume.
The default prefetch is 0, which means that RabbitMQ immediately dumps
all outstanding messages to the consumer, which slowly processes and
acknowledges them. If a second consumer were to connect to the same
queue, they would receive no messages to process, as the first
consumer has already been allocated them. If the first consumer
disconnects or crashes, all prior events sent to it are then made
available for other consumers on the queue.
The consumer does not know the total size of the queue -- merely how
many messages it has been handed.
No change is made to the prefetch here; however, future changes may
wish to limit the prefetch, either for memory-saving, or to allow
multiple consumers to work the same queue.
Rename the method to make clear that it only contains information
about the local queue in the consumer, not the full RabbitMQ queue.
Also include the waiting message count, which is used by the
`consume()` iterator for similar purpose to the pending events list.
Otherwise, if consume_func raised an exception for any reason *other*
than the alarm being fired, the still-pending alarm would have fired
later at some arbitrary point in the calling code.
We need two try…finally blocks in case the signal arrives just before
signal.alarm(0).
Signed-off-by: Anders Kaseorg <anders@zulip.com>
SIGALRM is the simplest way to set a specific maximum duration that
queue workers can take to handle a specific message. This only works
in non-threaded environments, however, as signal handlers are
per-process, not per-thread.
The MAX_CONSUME_SECONDS is set quite high, at 10s -- the longest
average worker consume time is embed_links, which hovers near 1s.
Since just knowing the recent mean does not give much information[1],
it is difficult to know how much variance is expected. As such, we
set the threshold to be such that only events which are significant
outliers will be timed out. This can be tuned downwards as more
statistics are gathered on the runtime of the workers.
The exception to this is DeferredWorker, which deals with quite-long
requests, and thus has no enforceable SLO.
[1] https://www.autodesk.com/research/publications/same-stats-different-graphs
Currently, drain_queue and json_drain_queue ack every message as it is
pulled off of the queue, until the queue is empty. This means that if
the consumer crashes between pulling a batch of messages off the
queue, and actually processing them, those messages will be
permanently lost. Sending an ACK on every message also results in a
significant amount lot of traffic to rabbitmq, with notable
performance implications.
Send a singular ACK after the processing has completed, by making
`drain_queue` into a contextmanager. Additionally, use the `multiple`
flag to ACK all of the messages at once -- or explicitly NACK the
messages if processing failed. Sending a NACK will re-queue them at
the front of the queue.
Performance of a no-op dequeue before this change:
```
$ ./manage.py queue_rate --count 50000 --batch
Purging queue...
Enqueue rate: 10847 / sec
Dequeue rate: 2479 / sec
```
Performance of a no-op dequeue after this change (a 25% increase):
```
$ ./manage.py queue_rate --count 50000 --batch
Purging queue...
Enqueue rate: 10752 / sec
Dequeue rate: 3079 / sec
```
This system can't update stats while the queue is idle, without using
threads for this, but at least we ensure to update the file after
consuming an event if more than MAX_SECONDS_BEFORE_UPDATE_STATS passed
since the last update, regardless of the number of iterations done so
far.
The race condition is described in the comment block removed by this
commit. This leaves room for another, remaining race condition
that should be virtually impossible, but nevertheless it seems
worthwhile to have it documented in the code, so we put a new comment
describing it.
As a final note, this is not a new race condition,
it was hypothetically possible with the old code as well.
Anders Kaseorg
Tim Abbott
Christopher Chong
Zixuan James Li
Alex Vandiver
Zixuan James Li
Sahil Batra
Mateusz Mandera
Lauryn Menard
Shlok Patel
shanukun
Abhijeet Prasad Bodas
sahil839
Cyril Pletinckx
Steve Howell