These scripts are somewhat rough but I needed them to help debug a memory
leak we have noticed in rails 6.
The biggest object script finds all the biggest objects we have in memory
after boot.
The test memory leak runs a very simple iteration through all multisites
and observed memory.
This updates rubyzip library so that callers can trust entries when
extracting files avoiding situations where a rogues zip imported by a rogue
admin could cause a disk space issue.
This corrects an XSS in ?pp=help.
Also removes the jQuery dependency from rack-mini-profiler and restricts
memory sensitive profiling methods development only.
Per new lifecycle https://developers.google.com/web/updates/2018/07/page-lifecycle-api
On Android and latest Chrome when an app transitions from "frozen" to
active the new "resume" event fires with no accompanying "visibilitychange"
event.
This means that often background tabs may be stuck thinking that discourse
has no focus when, indeed, it has.
This leads to cases where no posts are marked read anymore.
In Rails 6 due to internal changes, the following sequence no longer works:
```
RAILS_ENV=test bin/rake db:migrate
RAILS_ENV=test bin/rake db:schema:dump
dropdb discourse_test
createdb discourse_test
RAILS_ENV=test bin/rake db:schema:load
RAILS_ENV=test bin/rake db:migrate
```
What appears to be happening is that our tracking of plugin migrations is
being missed on schema:dump or load.
A more comprehensive fix restoring schema:dump / load support will be
investigated.
This renames the DISCOURSE_ENV_HOST var @eviltrout introduced in 95a9a544
to DISCOURSE_ENV_HOSTS and allows for a comma delimited list of hosts
This is useful for testing plugins and customized host names
This makes sure that all processes that fork off the master have a fully
operation schema cache.
In Rails 6, schema cache is now bolted to the connection pool. This change
ensures the cache on all pools is fully populated prior to forking.
The bolting of cache to connection pool does lead to some strange cases
where a connection can "steal" the cache from another connection, which
can cause stuff to possibly hang or deadlock. This change minimizes the risk
of this happening cause it is already primed.
We make a STRONG assumption that the schema is always the same on all sites
when we spin up a multisite cluster.
Prior to this change plugin migrations were not working and multisite
migrations not working.
Rails internals changed so we need to account for it.
Specifically semantics of `db:migrate` in rails changed so it is sort of
a "multisite:migrate".
This is a temporary workaround for the issue in https://github.com/rails/rails/pull/36949
Discussing a proper fix in Rails with the Rails team.
Prior to this fix we were spinning up a thread every time we closed a connection
to the db.
New version of mini scheduler allows you to select the name of a schedule
in the history page in `/sidekiq/scheduler/history`.
This is handy for quickly looking up timing trends.
Well all this does is amends a commit comment cause I was over eager and
pushed previous commit.
This is the comment I wanted....
This allows `pkill -USR2 -f 'ruby bin/unicorn'` to restart current unicorn
It is handy if you want to bind a keyboard shortcut for unicorn restarts
in dev.
By doing so you can avoid finding the terminal, hitting ctrl-c and then
hitting up, enter, heading back to browser.
It saves time.
Automate stuff, you will not regret it...
In dev mode sending USR2 to the unicorn master supervisor process will
restart unicorn.
This allows a simple script like this to restart unicorn in dev:
```
#!/usr/bin/env bash
kill -s USR2 `ps aux | grep ruby\ bin\/unicorn | grep -v grep | awk '{print $2}'`
```
This commit introduces 2 features:
1. DISCOURSE_COMPRESS_ANON_CACHE (true|false, default false): this allows
you to optionally compress the anon cache body entries in Redis, can be
useful for high load sites with Redis that lives on a separate server to
to webs
2. DISCOURSE_ANON_CACHE_STORE_THRESHOLD (default 2), only pop entries into
redis if we observe them more than N times. This avoids situations where
a crawler can walk a big pile of topics and store them all in Redis never
to be used. Our default anon cache time for topics is only 60 seconds. Anon
cache is in place to avoid the "slashdot" effect where a single topic is
hit by 100s of people in one minute.
Start tracking the date an api key was last used. This has already been
the case for user_api_keys.
This information can provide us with the ability to automatically expire
unused api keys after N days.
This allows custom plugins such as prometheus exporter to log how many
requests are stored in the anon cache vs used by the anon cache.
This metric allows us to fine tune cache behaviors
Prior to this commit if we ever checked in a per_host directive and scheduled
job was previously global, total chaos could ensue as a feedback loop would
start
We very rarely used per_host to date, but just started making use of it
for heartbeats
We refactored some code to avoid app event leak in tests at c462c2f2
Unfortunately this broke all background notifications in title and they
were broken for the past multiple months.
This reverts commit e805d44965.
We now have mechanisms in place to ensure heartbeat will always
be scheduled even if the scheduler is overloaded per: 098f938b
Under extreme load on large databases certain regular jobs can take quite
a while to run. We need to ensure we never starve a sidekiq from running
mini scheduler, cause without it we are unable to queue stuff such as
heartbeat jobs.
Databases can have a lot of user actions, self joining and running an
aggregate on millions of rows can be very costly
This optimisation will reduce the regular window of consistency down to 13
hours, this ensures the job runs much faster
In some very specific cases (large sites) shared drafts can introduce a
performance hit due to the mechanism used to filter out topics
This avoids the entire process when shared drafts are not enabled
This is a very long standing bug we had, if a plugin attempted to amend a
serializer core was not "correcting" the situation for all descendant classes
this often only showed up in production cause production eager loads serializers
prior to plugins amending them.
This is a critical fix for various plugins
This reverts commit 39c31a3d76.
Sorry about this, we have decided againse supporting 0-RTT directly in
core, this can be supported with similar hacks to this commit in a
plugin.
That said, we recommend against using a 0-RTT proxy for the Discourse
app due to inherit risk of replay attacks.
Every time we type a letter the composer issues a reply length check.
This is due to the interconnecting components, the title one depends on the
body which means that this decision making is passed along each time, even
if the title does not need it strictly.
Anyway...
This optimisation has 3 parts:
1. If the composer string is super long (10000 chars) we will bypass, quote
stripping and space squashing.
2. Quote stripping is now done much more efficiently, we strip them all in
one go
3. Space squashing eg: `hello world` to `hello world` is done in an
efficient loop to avoid needing to generate superflous strings that need
GC
This change allows themes and components access to theme assets.
This means that inside theme js you can now get the URL for an asset with:
```
settings.theme_uploads.name
```
This adds a 1 minute rate limit to all JS error reporting per IP. Previously
we would only use the global rate limit.
This also introduces DISCOURSE_ENABLE_JS_ERROR_REPORTING, if it is set to
false then no JS error reporting will be allowed on the site.
Serializer is injecting information into cooked and reaching direct to
custom fields that were not preloaded
This amends it so basic post serializer can use the proper interface
That said we should probably follow this up so we don't reach for this
info on every post.
Previously the default stack suppressor in rack-mini-profiler was excluding
the plugin directory.
This made islolating issues more complicated cause you needed to defer to
pp=full-backtrace which is both slow and noisy
Previously we relied on side effects to set tracking state correctly
when inviting groups to messages
Also has a minor optimisation in that we use pluck instead of pulling in
full record
Previously we would unconditionally issue an "invited_to_pm" notification
to all non muting users.
New behavior
- Watching and Watching first post get notified
- Tracking get a new "summary" message
- The rest get nothing
This is consistent with topic creation and way clearer
In some very rare cases CssParser could be loaded but CssParser::Parser not
this ensures we check for the actual constant we plan to call for concurrent
digest generations
Running this inline makes more sense otherwise there is extreme risk in
saturating sidekiq queue.
This also reworks ordering and selection so we double check if a post needs
rebaking prior to rebaking, this unlocks the ability to run this rake task
from multiple consoles.
The message_bus performs a fair amount of work prior to hijacking requests
this change ensures that if there is a situation where the server is flooded
message_bus will inform client to back off for 30 seconds + random(120 secs)
This back-off is ultra cheap and happens very early in the middleware.
It corrects a situation where a flood to message bus could cause the app
to become unresponsive
MessageBus update is here to ensure message_bus gem properly respects
Retry-After header and status 429.
Under normal state this code should never trigger, to disable raise the
value of DISCOURSE_REJECT_MESSAGE_BUS_QUEUE_SECONDS, default is to tell
message bus to go away if we are queueing for 100ms or longer
This feature allows @ mentions to prioritize showing members of a group who
have explicit permission to a category.
This makes it far easier to @ mention group member when composing topics in
categories where only the group has access.
For example:
If Sam, Jane an Joan have access to bugs category.
Then `@` will auto complete to (jane,joan,sam) ordered on last seen at
This feature works on new topics and existing topics. There is an explicit
exclusion of trust level 0,1,2 groups cause they get too big.
Previously we would only hold the post process mutex for 1 minute, that is
not enough when processing a post with lots of images. This raises the bar
to 10 minutes.
It also cleans up error reporting around distributed mutexes expiring. We
used to double report.
The global setting disable_search_queue_threshold
(DISCOURSE_DISABLE_SEARCH_QUEUE_THRESHOLD) which default to 1 second was
added.
This protection ensures that when the application is unable to keep up with
requests it will simply turn off search till it is not backed up.
To disable this protection set this to 0.
The behaviour of #TERM in search has been amended
1. We try category or subcategory slugs
2. We try tags
3. We try tag-groups
The term `hello #my-group` will search for all posts tagged with any of
the tags in the tag group `My Group`
Future work may be introducing a slug cache here or caching it in the table
but the assumption is that the number of tag groups will not be huge
Previously users were still allowed to create topic via API even if
uncategorized was disabled.
Not 100% happy with all this special casing, but I guess we have to do
something.
This also splits up a mega spec now that we have fab! into a more easy to
understand structure (I hope)
followup to #bc03c509
There were 2 problems
1. VALUES was not properly getting multiple results ... we need (1),(2),(3)
not (1,2,3)
2. changes was mistakenly lazy evaluated eg `changed ||=` which meant some
queries were not running
This fixes a condition where an intermittent db connection could cause
invalid site settings to be stored
It also removes a catch all we had.
Somewhere around Rails 5 `db:create` started wanting full environment
this is a problem for Discourse since it needs to boot up data from the
db.
This removes the catch all and surgically adds a db / redis bypass to
db:create task.
Sometimes we would like to create a base image without any DB access, this
assists in creating custom base images with custom plugins that already
includes `public/assets`
Following this change set you can run:
```
SPROCKETS_CONCURRENT=1 DONT_PRECOMPILE_CSS=1 SKIP_DB_AND_REDIS=1 RAILS_ENV=production bin/rake assets:precompile
```
Then it is straight forward to create a base image without needing a DB or
Redis.
We now treat any external_id of blank string (" " or " " or "", etc) or a
invalid word (none, nil, blank, null) - case insensitive - as invalid.
In this case the client will see "please contact admin" the logs will explain
the reason clearly.
This previously was a hot path in topic view. Avoids an expensive active
record operation and instead perform SQL directly which is far more
targeted and efficient
Sam Saffron