This cache should save 2 database queries whenever we send a private
message. However, previously it was per-process (which meant it was
mostly useless) and also buggy (it never stored anything in the cache,
so that it was completely useless). Switching this to our standard
memcached setup will address both problems.
(imported from commit 1d807f30704bccf28de33a80523488aedc58a9be)
This fixes a nondeterministic test failure for me.
The first message sent in the test suite appears to get dropped. I don't know
why this is, and I'm pretty sure it was an existing bug. This message used to
be the one disabling the tutorial, which might explain why that didn't always
work.
Regardless, this commit at least makes the test suite usable, and we can work
on fixing that bug later.
(imported from commit 063e40871b9883e3a6dab93a4e0a51c5b2dae4b7)
Previously we only used these caches for Tornado requests, because we
were not updating memcached when e.g. the user's pointer changed, and
so functions like update_pointer would not work correctly.
Now that we are updated memcached when the User and UserProfile
objects change, we can use these for all requests.
This saves 2 database queries on every Django request to the server.
(imported from commit aa5bffd885d14bde38b95e80a226bd5ab66f253d)
This was biting us before when the user would leave a narrow before a
get_old_messages call associated with it finished. Specifically,
search.maybe_highlight_message() would assume a message was in the
DOM when it wasn't any more.
We also have to hide the 'loading more messages' indicator when
reseting the 'load more' status because otherwise it wouldn't get
hidden like normal in the load_old_messages() continuation, causing a
load_more_messages() not to fire when re-entering a narrow.
(imported from commit 4a136dd01305b039c0970f897b07e603b87d5d8e)
This should substantially decrease the amount of server load generated
by the userpresence system.
I tested that this indeed was indeed saving one query on
/json/update_active_status requests on my laptop with 2 users from the
humbughq.com realm logged in.
(imported from commit 03e9d4eb95b9f664d489862684ae162db2076e08)
This cache filling code takes about 5 seconds to run, which means it
will finish before Tornado starts reading from this cache, but if that
were to change, it would be much better to have made at least some
progress.
(imported from commit 60a3420cdb9ddf331d83573a3fdb6be1a5ee5a4f)
Previously we were calling select_related() on Message.client, which
doesn't exist. It seems kinda poor that this doesn't raise an
exception.
I believe this issue was causing us to do very large numbers of
database queries during get_updates calls during server restarts.
(imported from commit b79bd698820fbd9dd82bd61fc175c32cd5ce6d05)
Previously, AdminEmailHandler would crash if the record did not
contain any exception information (e.g. because it came from
logging.error()).
(imported from commit 868ec68f3a57eddbb2c851ecf9894e5eddf21e4c)
Previously when we logged errors using logging.error() but didn't
throw an exception, we would send these totally useless error
humbugs/emails that look like this:
Error generated by Anonymous user (not logged in)
No stack trace available
This change makes those messages include the actual text passed to
logging.error(), which is substantially more useful.
(imported from commit 76a8220ffe66d19cb0ca8ba9d1b42d5ecc4fd511)
But discard any changes the Django response middleware may have made
to anything other than the content.
This allows us to, for example, output our nice database query logging
for get_updates requests.
(imported from commit e1d2fd38ceb4d73ff50bdfaad7c72ddb24d0fe16)
This is preparatory for running the Django response middleware on
our Tornado responses.
(imported from commit 05da8ea9cb663a928b2f98a928f3992aae4f067c)
If the user scrolls super fast, our scroll handler might not catch
the user passing by some messages.
(imported from commit 14cebffcd1321f02443971ac5e1c922db19648ab)
This will mainly be useful in the event system branch, where we want
to actually send a response from a file other than tornadoviews.py.
(imported from commit b7ae9bb9b062215ab44eb5f0a3a72d6baeee1d07)
Since we flush memcached when we do a server restart, the flurry of
get_updates requests that fly in afterwards are all cache misses for
getting the User/UserProfile objects, so Tornado ends up spending
around 70ms per get_updates request rather than the usual 1-2ms.
So this should substantially improve our Tornado performance around
server restarts.
(imported from commit 07b8126bdfd4ff14e4c3362f9eda1fe5fd571c5b)
Our previous code could in theory end up clearing the caches it had
just filled, if Tornado's cache filling work happened to be faster
than the memcached flush.
(imported from commit 48174aadad398fb7a7c917a1df765c1261b12a55)
We create a circular reference between handler functions and our
wrappers for them so that we can pass the wrapper to jQuery.off when
users pass the original handler to us. This reference-counting
system can't break all the circular references we create because
users can unbind event handlers without explicitly naming the
handlers they want to remove (they can remove all bindings on an
element, for example). For now, we hope that this memory leak isn't
too bad.
(imported from commit 9615b5761b4b09ca7ca52c0d847e9b83330373fa)
Previously, we couldn't actually unbind some event handlers. The
problem was that when a user called $.off(events, handler), the
passed handler wouldn't match any that were actually bound because
the handler that was actually bound was our wrapper.
This bug specifically caused the handlers for our idle timers to
never be unbound, effectively never cancelling them.
(imported from commit 48efac954994a05c356d326e64a78ab0ace9fe3e)
We will need this for removing event handlers. This will
unfortunately create a memory leak, but we'll partially deal with
that later.
(imported from commit e439cb44d245e16d2254d1be053b68015a1f4c79)
Sometimes Dropbox shares with /s/ and sometimes with /sh/,
and I'm not sure which controls it, but we should deal with both.
(imported from commit 2222450f25c418b5fbd60ab2c30477467e34c0d1)
Previously, if for some reason pointer updates were not returning from
the server, the client would resend its request every second, rather
than waiting for the previous request to fail before sending a new
one.
(imported from commit d134adc50aabd135c7631913fecab3519aca6640)
It's closer to a presence query than an update, and more importantly
this moves this out of Tornado -- previously Tornado was spending at
least 3ms per recipient on messages sent to the MIT realm fetching all
this data to return back to users. This should save around 100ms per
message sent to a popular stream the MIT realm -- but more
importantly, each such event is 100ms during which Tornado is not
processing other messages.
(imported from commit 134169f0fdcd9f6640fda957edc4a28b07783d8e)
This is required because our migration is going to go in two phases.
When we do the database migration (on pushing to master), we update
all messages at that point. But prod doesn't know about the new
flags field, so any new messages sent on prod will not have the
read bit set.
When we push to prod, we want to re-run the bit of the migration script
that automatically sets read flags on messages older than the users's
pointer.
(imported from commit 961d33e972eac9ada80089bf1b1269c7fb42d56b)
We also needed this when rerendering on append, so moving it into
_maybe_rerender allows the two places to share the code.
(imported from commit 027d99cae7864747cf1ec94c95e8ece495b5c907)
This avoids our repeatedly retrying to fetch a tweet that doesn't
exist from the Twitter API.
(imported from commit b4ca1060d03da21e7e59e5b99e682d2e8457df15)
It's pretty confusing if this doesn't change. In some other world we
could update the fade, but since we're currently only fading on reply,
I think it would be weird to update the fade when you're picking a new
recipient.
(imported from commit 8f77419d443d578068b57f847354ac6da7632ee2)
Previously, we compared the recipients of messages to the message that
you triggered the reply off of -- even if you did a reply-to-sender.
This commit changes the code to instead track what you faded by,
rather than just the message you faded on.
Fixes#1037.
(imported from commit d9e2cb4122501b1bc45e231d4b52c2e7f9284fdd)
Previously, if you renarrowed, all message fading would be cleared
until you close and then reopen the compose box.
Fixes#1024.
(imported from commit 57981ba29ab597c4c84ca6e4e9d04a8284f49117)
We treat these exceptions the same way we treat fatal errors: report
the error message to our server and then allow the exception to reach
the top level.
We could also override document.onerror, but don't. There are a
couple of ramifications of this:
* Exceptions caused by event handlers directly attached to DOM
elements aren't handled
* Exceptions caused by code at the top level that triggers an error
(such as parse errors in our Javascript files) aren't handled
The reason we don't override document.onerror is because the
document.onerror handler has a limited interface and doesn't receive
the exception object. It only gets the message, file, and line
number of the error. Additionally, exceptions that we allow to
propogate out of blueslip trigger an onerror event when they're never
caught. In order to avoid handling the error twice (once by blueslip
and once by the onerror handler), we'd have to encode the fact that
the error has already been handled in the error message, which is
pretty ugly.
(imported from commit 7f049ae519dc198a9f7cfd41fd5dd18e584bd061)
This is to let us pass in the stack trace of an existing exception,
which will be required in a upcoming commit.
(imported from commit 421366a7a01deb770b7620417fb4660769c5db53)
The referenced element where the error was supposed to go was removed
in 66fd42914e4fc33719c4f21ad401748989f20b49. Now the error message uses
the regular compose error message area.
(imported from commit c82a6d863fa327ba982157d0b0607545d7e65cb7)
Previously, if the pointer was high on the page such that there was a
lot of empty space below and the render window was full (a situation
we could get into if we're following new messages arriving), a newly
recieved message would not be rendered even though it would have been
visible on the screen if we had done a rerender. Adding the
_maybe_rerender() call to the end of append() ensures that we don't
end up in this situation.
(imported from commit 925d3cc62e8221b42f1d5ff1788e99c7d07ccc24)
Now that Zev's message list branch is merged, there's no longer a
performance penalty for loading these old messages, and it improves
our narrowing performance to have them loaded.
This code is slightly different from the original commit
93d47710891cfc4db9fa00beaa5ccd10113aa1c3 since the way to access the
first element in the message list and the API for get_old_messages
have changed since then.
(imported from commit f295f892bea9327eb8316225b7b98f0e3b3fdc9a)
This will hopefully make stream privacy more noticeable. We still don't
allow people to modify privacy after stream creation, however.
Since we now use a radio box on the stream creation modal we had to change
the selector used by subs.js to determine if a new stream was to be invite-
only.
(imported from commit 641a4fab74301a9b3ecd4b3859f010dd4ece193e)
I think my previous commit causes the tutorial to run more regularly
in the test suite, so we, in turn, need to be more systematic
about disabling it.
(imported from commit c3805438b0564874a358526d3592b86d147547c0)
We were previously having an issue where the tutorial could
be pre-empted if you got a few messages while you were first
logging in.
I have some reservations about this being slightly fragile, and a
better approach might be to just have a bit that we use to determine
whether or not you've already seen a tutorial. (Or potentially that
checks whether or not you've ever sent a message.)
(imported from commit f8858f64a36bcd25887b76314caff283929f340c)
The issue, prior to this fix, is that the presence list actually
gets drawn *on top of* our "more messages" popup, which is ugly.
(Toggling the z-index on either or both of them did not seem
necessarily to matter, but that's probably because z-index
is subtler than I understand.)
This fixes Trac #1078.
(imported from commit a255aadb1884cf6c659085b26a36d378f680e83e)
Tim Abbott
Jessica McKellar
Keegan McAllister
Waseem Daher
Zev Benjamin
Leo Franchi
Luke Faraone
Jeff Arnold