This is simply too time-intensive, at scale, to perform during the
critical period of an upgrade. The most performance-critical
codepaths (Tornado and presence) have moved to the narrow user cache,
which we continue to pre-fill.
This implied by its name that users could have more than one key.
They cannot, currently; make the code clearer by switching to the
explicit column access.
Only prefetching `realm` and `bot_owner` means that the cache values
filled by this process differ from the cache values that would be
stored on a cache miss -- leading to extra database queries for those
objects.
Use the same `base_get_user_queryset()` to ensure we get the same
related prefetches.
Because the main indexes on end_time either don't include realm_id or
do include subgroup, passing an explicit subgroup=None for
single-realm queries to read CountStats that don't use the subgroups
feature greatly improves the query plans.
This cache was only used in one place, which is infrequently
called (only when sending messages, or searching explicitly for a list
of users) and the overhead of maintaining the cache is not worth
trying to avoid the well-indexed lookup of the huddle.
We now fetch recipient object when querying "Huddle" object in
get_or_create_huddle_backend as this query is eventually used
to get the recipient object only in get_huddle_recipient.
This commit also updates the select_related call in the code to
populate Huddle objects in cache to pass "Recipient" as argument.
Previously no argument was passed to select_related and thus no
related objects were being fetched, with no non-null related fields
being present.
This commit updates code to pass "realm" and "bot_owner" args to
select_related call in get_users. We pass "realm" and "bot_owner"
args to get_users because the caches which this function is used
to populate are used for get_user and get_user_profile_by_api_key
functions and they also select both these fields when querying for
UserProfile objects.
We remove the cache functionality for the
get_realm_stream function, and we also change it to
return a thin Stream object (instead of calling
select_related with no arguments).
The main goal here is to remove code complexity, as we
have been prone to at least one caching validation bug
related to how Realm and UserGroup interact. That
particular bug was more theoretical than practical in
terms of its impact, to be clear.
Even if we were to be perfectly disciplined about only
caching thin stream objects and always making sure to
delete cache entries when stream data changed, we would
still be prone to ugly situations like having
transactions get rolled back before we delete the cache
entry. The do_deactivate_stream is a perfect example of
where we have to consider the best time to unset the
cache. If you unset it too early, then you are prone to
races where somebody else churns the cache right before
you update the database. If you set it too late, then
you can have an invalid entry after a rollback or
deadlock situation. If you just eliminate the cache as
a moving part, that whole debate is moot.
As the lack of test changes here indicates, we rarely
fetch streams by name any more in critical sections of
our code.
The one place where we fetch by name is in loading the
home page, but that is **only** when you specify a
stream name. And, of course, that only causes about an
extra millisecond of time.
This adds an assertion ensuring the type of `store` before accessing the
`cache_key` attribute that does not exist in the base class. Also note
that `.decode` returns `Dict[str, Any]` instead of a `str`.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
These changes are all independent of each other; I just didn’t feel
like making dozens of commits for them.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Fixes#2665.
Regenerated by tabbott with `lint --fix` after a rebase and change in
parameters.
Note from tabbott: In a few cases, this converts technical debt in the
form of unsorted imports into different technical debt in the form of
our largest files having very long, ugly import sequences at the
start. I expect this change will increase pressure for us to split
those files, which isn't a bad thing.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
During events such as stream / topic name edit for a topic, we were
running queries to db in loop for each message for reactions,
submessages and realm_id. This commit reduces the queries to be
done only for realm_id, which is yet to be fixed.
This is accomplished by building messages with empty reactions
and submessages and then updating them in the messages using bulk
queries.
A key part of this is the new helper, get_user_by_delivery_email. Its
verbose name is important for clarity; it should help avoid blind
copy-pasting of get_user (which we'll also want to rename).
Unfortunately, it requires detailed understanding of the context to
figure out which one to use; each is used in about half of call sites.
Another important note is that this PR doesn't migrate get_user calls
in the tests except where not doing so would cause the tests to fail.
This probably deserves a follow-up refactor to avoid bugs here.
Our recent change in 2fa77d9d54 to
disable the cached_db cache backend broke upgrade-zulip-from-git with
an attributeerror; we fix that by checking the session engine before
trying to access its cache-related attributes.
Now reading API keys from a user is done with the get_api_key wrapper
method, rather than directly fetching it from the user object.
Also, every place where an action should be done for each API key is now
using get_all_api_keys. This method returns for the moment a single-item
list, containing the specified user's API key.
This commit is the first step towards allowing users have multiple API
keys.
For realms that don't have any presence-active users, we know for a
fact that there aren't any active clients that will be reloading just
after the server restarts, so we can skip filling the cache with data
related to that realm.
For zulipchat.com, this results in a significant performance
optimization for the recipient and stream caches, and a moderate
performance improvement for the user caches as well.
Private messages make up the bulk of Recipient objects. While private
messages are ~50% of messages, if you weight by messages received
(which is what is important for message-loading performance), it's
pretty strongly balanced towards stream messages.
We don't need to include long-term idle or other inactive users here,
since fetching them consumed to vast majority of the time.
(On chat.zulip.org, this decreased the runtime for populating the user
cache by 5x, removing only users we're unlikely to need to access).
