Messages that get sent out when someone subscribes many people to a new stream each
cause individual database queries (and their associated transactions). With the patched
bulk_create (which sets the .id on created objects), we can reduce this query down to a constant
number of queries on the Message and UserMessage tables.
Note for deployment (local dev, staging and prod):
you must be running a patched django, found here: https://github.com/acrefoot/django/branches
use this branch: acrefoot-bulk_create_with_id-1.5.1
on acrefoot-bulk_create_with_id-1.5.1
relevant sha1: ac6d885b811f7e2e34f0db0da217983f7dfd357f
(imported from commit b0dab9dac784d3ff47751e65bf22c2dddc22edf5)
* Change the highlight colors for private and mention messages
* Put timestamp and message controls into a single line
* Modify layout to allow more flexibility in control placement
* Lighten narrowed view background color
* Adjust composition area columns
(imported from commit c7edca358b079da0ca76fa26d998946574bded6a)
We also record the historical edits to the message in this JSON format:
[{"prev_content": "new test message 14", "timestamp": 1369157249},
{"prev_content": "new test message 13", "timestamp": 1369157118}]
but we don't actually do anything with the information as of yet.
(imported from commit 2d5ca449b87b33ad035ab0e076a22e150c8e7267)
* Modify the narrow icon in FontAwesome to make it better align to the pixel grid and display well on Windows+Chrome.
* Move the message controls to the right
* Hide the message info icon until the message is hovered / selected
* Switch the star to a gray version
* Increase the size of the gravatar
* Adjust the spacing
* Add the right-side message pointer
* Fix private message background colors and mention colors
* Modify star count test to account for new stars
* Bug fixes for stream subscription messages and other miscellanea.
(imported from commit 3d3d9de7e03f3658c5c78b492051b2b7f795487d)
This goes back to only scrolling by the size of the new
message, and it avoids scrolling in certain use cases.
(imported from commit f9e6380b779bb21283ba889715712b6b51633838)
Previously, we were referencing the mixpanel objects only once, at
page load time, which meant that there was a race between metrics.js
loading and mixpanel completely loading. Mixpanel starts with stub
methods and then replaces them once it fully loads, asynchronously.
If metrics.js ran before mixpanel loaded, we'd end up wrapping the
stub methods instead of the real versions. Adding a layer of
indirection ensures that we always get the right method.
(imported from commit 6a8cfbf249168443956895b7a7e29bf7bb4222aa)
I apparently screwed up my check for whether we were using the old
data-name field, and switching the other stuff to use data-id (which
is needed for the color stuff) is probably not worth it.
(imported from commit 1b925bbcca5beb5dc9dadbcf703cbb07ca511a0c)
I think they look a lot better when sized so that the
Subscribe/Unsubscribe button and the labels on the left are both
centered within their respective rows (and also within the blue
regions that hovering over the row displays), and this seems to cause
that to happen within a wide range of font sizes.
(imported from commit d586aecee4b16540ad480509b5b888bd8de02cf0)
There was no benefit to our various link processors all doing
independent scans through the list of messages, and this makes it much
easier to understand the logic of how each link will be handled, and
also makes policies like "don't process links if there are more than 5
of then" easier to implement coherently.
(imported from commit 4affdeab889ba89b99eec905fdf871e78bbc3dd4)
This reverts commit 87226d857845c6f16cb3bc0d6ab5bb748aca5987.
This meant that if for some reason there's a server error or network
failure trying to send in your edit, your changes are silently lost.
(imported from commit 2b5d19716fef1565b061a2b6c7cecc54f183b6f3)
It's not as clear as it could be which stream you clicked on from the
location of the popovers, so it's worth making the popover clear about
which stream you're modifying.
(imported from commit 289b2e70eab582f4ec12d62410e095fd632f6582)
Currently, some browsers don't seem to be sending metrics information
to mixpanel. This commit will make said browsers noisy, but should
help debug what's going on.
(imported from commit c5050f66d985eb76e38117b2668594fedfc10702)
Still not perfect, but now we move the pointer down and scroll
to make sure that the newly read messages are truly marked as read
(imported from commit 2b9a14d1c8695eac0ed9fb03484068dd9b08b940)
Constrain the meat of the page to the center 1440px or so.
This is achieved in a slightly more hackish way than I'd like,
but I think it's mostly necessary if you want the long color
bars that extend beyond that main area.
(I encourage you to view this diff with -w)
(imported from commit 10bf4462411146090b0147218d51cc444c3c91a2)
Use tab_bar_underpadding to find out the top of viewport
that we can view. Also, eliminate effective_page_size().
(imported from commit 0e2d777790552e77d635989e496f3446cefccb1e)
We're still failing frontend tests randomly due to this timeout being
too short. Ultimately we should fix this by making the
wait_for_receive check smarter, but this will do for now.
(imported from commit ff4b18beb88b957c705fd98cd9064902c9985f62)
Currently our test database is backed by sqlite; this commit moves
us to using postgres for our all database needs. This, in conjunction
with the patched django on github, allow us to have fewer hacks and
more true-to-life tests. It also sets the stage for testing the bulk_create
and schema search_path patches made to django.
Developers will need to run:
./tools/postgres-init-test-db
./tools/do-destroy-rebuild-test-database
this is assuming that they have already run:
./tools/postgres-init-db
./tools/do-destroy-rebuild-database
at some point on this pg_cluster. (The ordering is important; it will other-
wise complain about the south_migration table).
(imported from commit c56c6f27e13df7ae10b2e643e65d669dde61af3d)
This will make it automatically work if we add new tab
bar like things. The current version of ui.message_viewport_info()
is slightly broken; that's a separate fix.
(imported from commit fa1906b738433223831250e3191dfd8e87d67daf)
Most of the model logic pertaining to unread counts had been in
zephyr.js, along with a couple global variables. Now the code
is encapsulated in unread.js. It was a pretty straightforward
extraction with some minor method name changes. Also, a small
bit of the logic had also been in stream_list.js.
Conflicts:
tools/jslint/check-all.js
(imported from commit f0abdd48f26ab20c5beaef203479eb5a70dacfff)
Set the background behind "Private messages" to green whenever
a user's unread count goes up for private messages. Remove
the background after 3s. Advanced browsers will fade the
green in and out over 6s (3s up, 3s down).
(imported from commit 80ed9661d9eec1d697f3259854037d7e145615cd)
The only known outstanding bug with this is that it doesn't properly
handle the updating of a message's highlighting/presence in a narrowed
view (e.g. in theory, a message should disappear if it is edited such
that its subject doesn't match your narrow or it no longer matches
your search). I think I'll just open a trac ticket about that once
this is merged, since it's a little hairy to deal with and kinda a
marginal use case.
Also it's not pretty, but that should be easy to tweak once we get the
framework merged.
Conflicts:
tools/jslint/check-all.js
(imported from commit 2d0e3a440bcd885546bd8e28aff97bf379649950)
Currently the interface for editing messages is limited to a
command-line API tool; it's great for testing with e.g.:
./api/examples/edit-message --message=348135 --content="test $(date +%s)" --site=http://localhost:9991 --subject="test"
The next commit will add a user interface for actually doing the editing.
(imported from commit bdd408cec2946f31c2292e44f724f96ed5938791)
Specifically:
* Leave the avatar image as inline and round it.
* Move timestamp to the left column.
* Replace the "Info" link with a permanent info sign.
* Move the pointer bar to the left.
* Remove borders
* Change selection background colors, and PM colors.
* Introduce the "narrowing" icon into our FontAwesome set.
* Modify the tests to account for the new "narrowing" icon and fixed a bug in star-finding.
* Clean up CSS and add a more prominent color to private messages
(imported from commit 8a8d6de8acccc52c0d16f5d1ce31aabdc72c88c8)
For beanstalk we need to provide a decorator that converts %40 to @ in the
http basic auth part of the URL. However, if we put our own wrapper around
rest_dispatch, the Django CSRF protection jumps in. This requires us to put
@csrf_exempt on our extra dispatch function, at which point we might as well
have avoided rest_dispatch in the first place and put a @csrf_exempt decorator
on our api_beanstalk_webhook.
(imported from commit b1f459aad26a5b80cce93f6c859240a53c11cc22)
The geometry used by within_viewport() is a little off to
begin with, but we don't want to check it at all in this
situation, because if the last messages falls out of the
viewport, we still want to scroll. The relevant thing
to check is that available_space_for_scroll exceeds zero.
(imported from commit a0a6f0d23db2eab8d9f22fc9ad523031cf7f7ec2)
The prior code was subtracting out the compose box from the
calculation of available_place_for_scroll, which didn't make
sense when the compose box is at the bottom of the screen
and you're scrolling the current message up. You could see
the symptom pretty clearly by seeing autoscroll stop exactly
the height of the compose box from the top.
(imported from commit cfceb85c8be80cca957ac4a3ad0bbf0de7425c48)
Otherwise the stars and info icons are no longer green.
This reverts commit f0571fc9e3005a4f2975174f230f77fed17adcfd.
(imported from commit fdf3c54dbba53917fe300ddbb641408edaddc44f)
This fixes a pretty subtle bug where the window-focus handler
wasn't updating the unread counts in the title, but it was
hard to notice, because as soon as you moved the mouse, the
problem fixed itself.
Apart from fixing the bug, this patch eliminates the expensive
mouseover handler, which is a big win.
The fix to the window-focus involved some unrelated cleanup. I
decoupled update_title_count() from received_messages(), as the
former method will probably live somewhere else soon.
Also, in order to get window-focus to update the title count,
I went pretty deep in the stack and added a call to
update_title_count() inside of update_unread_counts(). This
fixes window-focus as well as restoring that behavior to
code paths that were calling received_messages().
You'll see that call to update_title_count() is now adjacent
to the call to update_dom_with_unread_counts(), which is
fairly sensible, but then are calls to similar methods like
notifications.received_messages() that happen higher up in
the call chain, which seems kind of inconsistent to me. I
also don't like the fact that you have to go through a
mostly model-based function to get to view-based stuff, so
there are some refactorings coming.
(imported from commit 2261450f205f1aa81d30194b371a1c5ac6a7bdec)
Nicer URLs:
/login to login
/register to register
/signup to signup
(Step two is to remove, e.g. /accounts/home/)
Also make /login the default page when you're not logged in.
(Prior to this commit, it was annoyingly different on deployed
vs. not.)
(imported from commit 21adb7a94f03256098d15b2e608d793d3ddb5b23)
Windows, Mac, iPhone, Browser icons from http://www.endlessicons.com/ and modified
Macbook Air image from http://psdsonar.com/macbook-air-free-psd/ and modified
Linux icon from Wikimedia Commons and modified
Android icon from Wikimedia Commons and modified
(imported from commit 3cf8617cf49a833b706a2ff78b986e28c21e26cc)
South doesn't properly deal with removing the Django User model, so
this commit redoes our South history to instead start after that
migration has already been applied. This allows us to get rid of some
annoying hacks.
Note that developers and staging will need to run
./manage.py migrate --delete-ghost-migrations zephyr
in order to clear out the old versions of the migrations.
(imported from commit 7f45ea601b809dde33720f76e7dfb0ab348b0e65)
This, combined with acrefoot's work on sending the notifications in
bulk, resolves trac #1142 -- we do only 10 database queries and the
whole operation completes in about 300ms on my laptop.
(imported from commit 36b5bb836bc6c713903d1ca72e39af87775dc469)
I renamed set_count_internal to update_count_in_dom, because "internal"
was redundant in terms of saying the function was private, and it misled
me into thinking it was internal-only in impact, but it actually updates
the DOM.
I also removed the synchronous callback functions, since they both
led to simply hiding the count_span and clearing the text of the
value_span.
(imported from commit dea27d6414dc1b33818b24662f8246d687530b71)
This allows us to load our own code before most dependencies are
loaded. Our compiled Handlebars files still need the Handlebars
runtime, so we can't move all of our minified code before
dependencies yet.
(imported from commit e2d0fa13f05a08fc3c2519790f7382e5eef6eca2)
The problem is that if you load a browser window in a stream narrow,
add_message_metadata will be called for the messages in the narrowed
view before it is called for the messages going into the main view
(thus inserting them into all_msg_list), resulting in duplicate
copies of messages.
This would be mostly OK except that we call
process_message_for_recent_subjects inside add_message_metadata, and
that function assumes it is only called once on each message
(otherwise it'll double-count the message).
(imported from commit a3e7f85874100cd93a6d07684605da04d9cc80c7)
Created a function message_viewport_info() to return more accurate
effective viewport info and called it from process_visible_unread_messages().
Also killed off a tiny bit of dead code in process_visible_unread_messages().
(imported from commit 985fcf2fb447dbf1026e2de37574c255a9bd6196)
Whenever we get a narrowing event, it's possible for new messages
to appear visible, and we need to call process_visible_unread_messages().
This has been a bug, but it's mostly obscured by the fact that we
call process_visible_unread_messages() as part of focus/scrolling
events.
(imported from commit b9447977f8e2272d45865ca67b436cacafd58a03)
This does the simple thing, which will work as long as the phrase
doesn't have any punctuation and contains the exact text being
searched for. We really want phrase search (check if the lexemes in
the query occur in sequence in the input lexeme string), but Postgres
doesn't support that natively.
(imported from commit 67bf36883ed21743fcad3f02ad5b319ab188f816)
This works fine on prod, but after a new install-server, the closure
compiler complains about a side-effect-free for loop init.
(imported from commit aa0e4d788abe4c819d4d912d6a526fab4f676675)
This removes the large "New stream message" and "New private message"
from the left sidebar. It also makes the default action when clicking
inside the composebox the same as the "New stream message" button used to
do (instead of replying to the stream-subject pair at the current cursor).
(imported from commit 316f03a35b781aca4c42555f74b99c4332ff42de)
THE CENSUS OF HUMBUG
Now it happened that at this time Waseem Daher issued a decree that
a census should be made of the whole users of Humbug.
This census -- the first -- took place while Faraone was governor of
MailChimp, and everyone went to be registered, each to his own realm.
So Alice Humbugger set out from the town of MIT for MailChimp in order to
be registered.
(imported from commit fca7714ebffd0b39b9b1337058f67975985f4039)
This refactoring basically splits off two functions from update_unread_counts(),
which then becomes a simple three-liner.
The function get_unread_counts() is extracted, and it's purely functional
computation. It paves the way for a more pull-based approach to getting "unread"
counts, where other parts of the program can just call it to get values as
needed without worrying about side effects. It is staying in zephyr.js for
now.
The other function is stream_list.update_dom_with_unread_counts(), which
has a new home in stream_list.js. It handles all the DOM manipulation
aspect of unread counts in the left pane, mostly by delegating to smaller
functions within stream_list. Some of those smaller functions can now
be turned into private methods FWIW, but I'm not sure it's worth the
trouble.
(imported from commit 799f9ebbaed8d530829a4741ef14be04bd8abf5a)
This is a prefactoring to eventually eliminate the home_unread_messages
global variable. More commits to follow.
In order to set up process_loaded_for_unread() not to modify
global variable to get its job done, we want to pull it out of
add_messages(), so that add_messages() doesn't have to pass back
state to the 9 different places in the codebase where it's called.
There are only 2 places where process_loaded_for_unread() get
called after this commit.
In order to facilitate pulling up process_loaded_for_unread(), I
made it so that the contract for add_messages() was to accept
already-hydrated messages. This way I could hydrate the messages
before calling process_loaded_for_unread() without have to
worry about double-caching them in add_messages. This will
slightly improve performance, but it was mostly done for code
clarity.
(imported from commit ad5aaad5b1f22c31647370f4c9dcb5f89d7d99a7)
This was a workaround for a number of other bugs we had, but at this
point just serves to make debugging more difficult.
(imported from commit 6662b7854c265bd8016f6c8ce75a095731211a45)
Since we log to statsd our cache time lookups by cache key, using a unique
tweet id for each lookup was just filling up our cache without being useful.
Also, log database cache lookups in a further namespace to distinguish between
memcached caches
(imported from commit a2a16b777fb7ab8cd066feee7344f9c8a3c107f5)
Users can send to any stream except invite-only streams that they
aren't subscribed to. Bots can send to any stream except invite-only
streams that neither they nor their owner is subscribed to.
(imported from commit 623d34d249d923611ca7ca781b5b55205cd3e548)
We really should not be storing bot API keys in the DOM and should
require some sort of additional authentication before showing them,
but this seems reasonable for a first pass.
(imported from commit c7d75aa52e21894bf53917457e771c18de38bbcc)
Previously a default was missing from this function, which resulted in
users being unable to change their settings if they tried to disable
sounds.
(imported from commit 2dae67dcb2e8cb986abb6dee9659be2192993dd9)
After this change, the memcached time consumed by doing
get_old_messages for 200 and 1000 messages respectively now look like
this:
200 63ms (mem: 6ms/3) (db: 4ms/2q) /json/get_old_messages
200 178ms (mem: 67ms/2) (db: 6ms/1q) /json/get_old_messages
which might help explain where the time is going on prod for some of
our slower queries.
(imported from commit b8fe83b175914b6796922a65a1c5537f4e7a9429)
The idea here being: if there's only one line, it discourages
me from writing a long message (and also makes me think that
enter will send).
(imported from commit 424d8d305d1965ce3199ce3227dac94b395945bc)
This commit takes control of keyboard-based pagination away
from the browser, so that we can use the effective viewport
size as the amount to page, as well as keeping a little bit
of overlap from page to page. There had been issues with
pagination for a while, but the introduction of the always-open
composebox particularly aggravated the situation.
(imported from commit 45b9b7d5a6b322230c9d55e1be0b763dbce06e2e)
For sites that are supported, we now grab thumbnails for images + video
embed code for videos and use them in lieu of our existing embed code.
We also embed rich non-script content.
Special casing is done so that we don't embed images twice.
Some testcases were modified to avoid triggering Embed.ly
The manual step is to install python-embedly.
(imported from commit d725bab91675c61953116c5ca741055fce49724e)
Previously we never sent desktop notifications when the browser was
focused, even if the message appeared offscreen. After this commit
there are only a few cases when PM or other notifiable message doesn't
trigger a desktop notification:
(1) You sent it yourself
(2) It was onscreen when it arrived while your Humbug window had focus
(imported from commit e381c02c0e6794594d6934f57249a11ba2a88210)
This is trying to make the logic flow clear -- e.g. we check once, at
the beginning, for whether the message is notifiable, and the checks
for whether the various notification settings are enabled are more
parallel.
(imported from commit a68c71a53055191bc16682a85f739ed8e40ddeae)
It's strictly more functional, and having a single arguments
extraction decorator makes our codebase less confusing.
(imported from commit 2a5618c04b486268a462a24a1481ac030f15eac4)
See #1234 for details. When you upload files the old-school way
(no drag&drop), there was a bug where you couldn't upload the same
file twice, due to us intercepting the change event and not clearing
out the file list when we were done. Tested on Chrome, but uses
a known IE workaround.
(imported from commit 8120c2e8bce41f3964f4f5c21aad3a85df0e433d)
The filedrop library has a few canned errors, but it seems to mostly
let server errors come through. We try to trap 413s to give a more
descriptive error than "unknown," but this is just a bandaid fix,
and we should see what's wrong with our prod configuration.
(imported from commit eac26406866d80340f24dbdca9f34408ddb92462)
The .height() and .width() functions are actually pretty expensive for
the number of times we call them. The viewport height and width
don't change often, though, so we can just cache them and recalculate
them on window resize.
(imported from commit 129fb8c058144125e2974f6b7967cd9f1a5c9ead)
The .height() and .width() functions are actually pretty expensive
for the number of times we call them. The callers of within_viewport
already know the offset and height of the row, so we just pass them
in so the values don't have to be recalculated.
(imported from commit d1c077bd87463d695f0bbe337b6a8b04ac2d17ce)
The optimizations are:
* Sort over the list of subscriptions instead of the DOM li elements.
This requires storing the li elements for each sub on the sub object.
* Do a bulk insert of the li elements instead of doing them one by one.
(imported from commit 1a987799930fc677e25f0bc2dcf66f83a4ac3163)
We now fire three events:
* subscription_add_done - fired when subs.js has finished handling a
subscription_add event (all structures are set up, etc.)
* subscription_remove_done - fired when subs.js has finished handling a
subscription_remove event
* sub_obj_created - fired when subs.js has created a sub object. This
happens both when a new subscription is added and at page startup for
all existing subscriptions
These events are fired whenever sub objects are created, even when
not tied to a subscription event.
(imported from commit a4863451f37e7fdbad480696b388ea788b01d6b9)
* Start a compose when we do a file upload
* Restore the "Formatting" and "Feedback" links.
* Dismiss composebox error messages when we defocus composebox
Realistically, the "correct" way to do this is not to have to
explicitly manage the composebox's state, as we do now -- it should
just be 100% visible and ready to send any time you click 'send'; it
shouldn't need to have first been composebox.start()ed.
(imported from commit 7f1725c229ed968a9b5500b25d600306173182a0)
What changed:
* Vector icons swapped in for the left sidebar buttons and filters
* Lighter font weight in the stream filters list
* Round color swatches in the stream filters list, with an inner shadow
* Tighter line height in the individual messages in the message pane
* Fixed button widths in the left sidebar (so the buttons are equal in width)
(imported from commit 337dc4a3d8e29945cfc8cfb9524ac76a7b038ad8)
Recently the typeahead for streams in the compose box was modified so
that streams only matched queries when the query was a prefix to the
stream. When that change was made, the old highlighting behavior
had been mistakenly left in place. This commit fixes the highlighting.
(imported from commit b7ec33daba46978df58eb91306686a4f1a57c7fa)
* Properly resize compose area when we cancel out of it.
* Re-enable clicking on 'reply' in popover.
(The issue with the latter is that clicking on "Reply" started
a reply and then bubbled up and triggered our code that canceled
a reply because you clicked out of the composebox.)
(imported from commit 25d0ea58b72d2ee246217baf3eb9cac58fc858f5)
Really, the "correct" way to do this is to undo "scrolltheworld", and
then just have a compose div that always lives underneath the message
list div. (This will also allow us to deal much more reasonably with
the whole "Is the composebox in focus" thing.)
In the interest of prototyping something more rapidly, though, we
adopt the somewhat more hackish approach, with the understanding that
much of it will probably be simplified later.
(imported from commit e2754be155c522b6dac28e7b84c62bd2030217c8)
We previously kept the lists in the DOM for all streams and updated
them all when new messages arrived. This was very expensive for
large numbers of streams, so we now just build the subject lists on
demand.
(imported from commit 937ad4322ce2014200aeae8645f79875f6af576e)
This commit also fixes a bug where "starred messages" wouldn't get
bolded when you narrowed to starred messages. However, it also
introduces a regression where subjects aren't highlighted correctly
on load to a narrow which will be fixed shortly.
(imported from commit 411575d92762e41d04c1baf126c0ab1dfb4225a5)
This will matter shortly as hashchange.initialize can call
narrow.activate(), which fires an event handler.
Really, I have no idea why we have these initialize() methods anyway
and we don't just do initialization on document.ready.
(imported from commit 3a6a80e1426b03439b95cae3f142a4b1c43125e9)
We memoize add_message_metadata by checking if the message is already
in the all_msg_list. Therefore, we need to add messages to that
message list before we add it to the narrowed_msg_list.
(imported from commit 4346179376ef6f982162c02c6152a0d294bfb2c0)
The String.localeCompare function is really slow, at least partially
because it creates a locale-aware collator object each time. So now,
when we can, we create and cache a locale-aware collator object.
However, this is not supported on most browsers, so we fall back to a
non-locale-aware comparison. This is not ideal, but for now we are
mostly working with English-speaking customers.
(imported from commit 51aa02e3b9fe4a0ef0cb084874fe26e91c57f65e)
Addionally, print out a blueslip error instead of dying
if a stream id is accessed when there is no stream to get
(imported from commit 0d6466ca79312a4fb9a235f313303ac5246afb35)
This decouples from Chrome notifications, which gives us cross-platform
support in at least modern browsers.
We log this action so its replayable in our message logs.
This implements the model change indicated by the previous schema commit.
(imported from commit b21213cdde54f43670bbb0bf1f607147fc732b38)
We test if the user supports sound in their browser, then determine which
sort of sound their browser supports.
When, whenever we show a desktop notification we also play a sound.
(imported from commit dae41e70a6e4f6ed60ffedaac546d77baee52675)
Since they can't be parsed, probably the best thing to do is to send
the user to the home tab; we could add in showing an error message but
then we'd need a way to clear the error message -- better to just have
this work.
(imported from commit 67c0475ff06eb0431621eef60b9c50287a158232)
Previously, we were fetching Message.objects.select_related() from the
database, even if we actually ended up fetching the message dicts from
memcached and thus not actually using them. Especially in the cached
case, this resulted in a lot of overhead where the Django ORM put
together Message objects with lots of data in them that were never
used. This commit switches the model to only fetch the full message
objects from the database for those messages which are not found in
the memcached caches.
Here are the timings for get_old_messages before this patch was applied:
(cached)
127ms (db: 42ms/2q) /json/get_old_messages (starnine@mit.edu via website)
385ms (db: 105ms/1q) /json/get_old_messages (starnine@mit.edu via website)
(uncached)
315ms (mem: 6ms/41) (db: 90ms/22q) /json/get_old_messages (starnine@mit.edu via website)
507ms (db: 94ms/14q) /json/get_old_messages (starnine@mit.edu via website)
Here are the timings for get_old_messages after this patch was applied:
(cached)
80ms (db: 9ms/2q) /json/get_old_messages (starnine@mit.edu via website)
133ms (db: 4ms/1q) /json/get_old_messages (starnine@mit.edu via website)
(uncached)
230ms (mem: 9ms/41) (db: 48ms/23q) /json/get_old_messages (starnine@mit.edu via website)
385ms (db: 55ms/15q) /json/get_old_messages (starnine@mit.edu via website)
(imported from commit c4748513392a906393314aa7cd41d98a69865411)
The .data() method tries to coerce the value of the attribute into a
Javascript type, which is not what we want when the stream name looks
like a number or some other Javascript type.
(imported from commit a5f639d2ef98435cec6beacf3837fc185474a955)
On page load, the scroll_finished function was being called and
scroll_start_message was -1. This caused us to mark all messages
that we loaded through the messages initially visible as read. This
was particularly problematic because message_range iterates over all
message ids between its two arguments.
(imported from commit d93209d466797939cc9dbdbe76d25a5b20195bd2)
Previously we were doing quadratic work in the number of streams
because we had to iterate over all <li> elements every time we added
a new one.
(imported from commit 60cb97f77d161e9d8c3072157fa9c57c58f7af52)
Since we pick a new color every time we add a new subscription and
recomputing the available colors was linear in the number of
subscriptions, we were doing quadratic work on page load.
(imported from commit 647ff3cb82f405755711da47701f005e7bc0023e)
We were previously doing this on every message. Because
update_recent_subjects is linear in the number of streams in the
sidebar, this became very slow when we enabled the streams sidebar
for the MIT realm.
(imported from commit 95cd71d83bbcc08cc6c5c79ca567b5d6b9b17173)
We were previously calling sort_narrow_list after each stream was was
added. Because it is linear in the current length of the sidebar
list, we were doing quadratic work on page load. When we enabled the
streams sidebar on the MIT realm, this became problematic because of
the number of subscriptions Zephyr users have.
(imported from commit d60ddc638f0a81fbce08eecd6671e9ea6ca38515)
Messages are now explicitly condensed by our JS, which means that if
we run into some bug where our JS doesn't run, you still see the whole
message (rather than getting a clipped message).
(As of this commit, this can happen when you, e.g. are on the
Settings page and someone sends you a message.)
(imported from commit f3bec97800ea1852c80203e73552ee545fcc7e8a)
This fixes a bug where if you were narrowed to a search and received
a new message that belonged in that search, the message would appear
to have an empty subject and content.
(imported from commit fe1dbf584d3659d57c5b70c7eb45cb22bbc9732f)
Previously, we were having this problem where:
* You narrow to something
* That causes message_list.js:process_collapsing to run on all of the
elements in the view, which changes some of their sizes
* That causes the pane to scroll and either push the content up or
down, depending (since stuff on top of where you were is now a
different size)
* That triggers keep_pointer_in_view, which moves your pointer
Moving process_collapsing into narrow.activate doesn't obviously
fix any of this, but it does seem to mitigate the issue a bit.
In particular, we (a) process it less frequently, and (b) process it
immediately after we show the narrowed view table, which seems to
reduce the raciness of the overall experience.
This does, however, introduce a regression:
* If you receive a long message when you're on
#settings, e.g., and then go back to Home,
the message does not properly get a [More] appended
to it.
(imported from commit b1440d656cc7b71eca8af736f2f7b3aa7e0cca14)
This can be useful for debugging what sort of narrow is happening in
addition to the URI decoding bug we're currently experiencing.
(imported from commit 0cb55fec4ac1afa986c747eb79236b4300c9e636)
This shouldn't have any effect in normal realms, but for realms like
mit.edu that have large numbers of inactive streams, it will sort all
the streams that have had a recent message at the top (aka those that
aren't effectively inactive).
(imported from commit 027ce258d04b6fd58705e49f769dec7e0639bb38)
We HTML-escape the subject in Postgres to avoid a server round-trip.
Unlike the rendered_content, which is already escaped and cached on
zephyr_message, we normally escape subjects client-side. Escaping in
Django would require fetching the messages that match the query,
escaping the subjects, and then making a second query to Postgres to
insert the markup. We could instead fetch the messages with subjects
marked up using non-HTML (some unique string) that is later converted
into the correct markup either in Django or client-side, but then the
escaping problem would just be with some random string instead of
HTML. Since the function is pretty simple, doing the escaping in
Postgres itself is the least painful option.
(imported from commit 004931d8e496697c18650aee97b1a74c55a04cb2)
In addition to changing the trigger that updates
zephyr_message.search_tsvector to use our new text search
configuration, it also now builds the tsvector on rendered_content
instead of content and fires on update of only the subject or
rendered_content columns.
This migration is expected to take a long time. The
checkpoint_segments parameter in postgresql.conf should be
temporarily raised (probably to 32) while it is running.
(imported from commit 4535438bb33ce1db2a74ecbe91efc52afdb568f1)
Text search was not that great partially because Postgres wasn't
using a ispell dictionary (Postgres term) before. We now pull in
Hunspell and use its dictionary and affix rules.
It is Ok to run with this new configuration before updating our full
text column and index that will be coming in the next few commits.
Manual steps for deploy:
1) On both postgres0 and postgres1 (both before moving on to step 2),
install the hunspell-en-us package
2) On staging, run migration 0022
3) On both postgres0 and postgres1, copy the appropriate postgresql.conf
file over
4) On both postgres0 and postgres1, run `pg_ctlcluster 9.1 main reload`
(imported from commit 706bf0f6ecc46c712cea10b73c34fd9d1dfd4767)
There's still a lot to do here. For example, the external code
should probably go through the new Filter object directly instead of
indirectly through the narrow module.
(imported from commit 22dcd31cdebd51453f1658af52a4432b2fe7a4cb)
In the case where we're getting old messages for a narrowed view, the
anchor message id might not actually be in the result set so there's
no reason to fetch an extra message.
(imported from commit e610d1f2cb95be3ff9fce6dc95e40c560bc5bf84)
In particular, I added absolute positioning and hidden overflow,
which ensures that if an element has a persistent min-width
(like a file input field apparently does), it doesn't affect its
parent.
(imported from commit 72e7a5bee2775fb6f229899ba849292eee76aa4a)
In repeated trials, the initial data fetch used to take about 1100ms.
In practice, it was often taking >2000ms, probably due to caching
effects. This commit cuts the time down to about 300ms in repeated
trials.
Note that the semantics are changed slightly in that we may no longer
get exactly 25000 messages. However, holes in the message_id
sequence are currently very rare or non-existent so this shouldn't be
a problem and we don't care about the exact number of messages
anyway.
I believe the problem was that the query planner was unable to
effectively use the LIMIT clause to figure out that only a small
subset of zephyr_message was going to be needed. Thus, it planned
for operating on the entire table and decided it could not use a more
efficient plan because work_mem, although large, would not be large
enough to execute the query over all of zephyr_message.
The original query was:
SELECT "zephyr_message"."id", "zephyr_message"."sender_id", "zephyr_message"."recipient_id", "zephyr_message"."subject", "zephyr_message"."content", "zephyr_message"."rendered_content", "zephyr_message"."rendered_content_version", "zephyr_message"."pub_date", "zephyr_message"."sending_client_id", "zephyr_userprofile"."id", "zephyr_userprofile"."password", "zephyr_userprofile"."last_login", "zephyr_userprofile"."email", "zephyr_userprofile"."is_staff", "zephyr_userprofile"."is_active", "zephyr_userprofile"."date_joined", "zephyr_userprofile"."full_name", "zephyr_userprofile"."short_name", "zephyr_userprofile"."pointer", "zephyr_userprofile"."last_pointer_updater", "zephyr_userprofile"."realm_id", "zephyr_userprofile"."api_key", "zephyr_userprofile"."enable_desktop_notifications", "zephyr_userprofile"."enter_sends", "zephyr_userprofile"."tutorial_status", "zephyr_realm"."id", "zephyr_realm"."domain", "zephyr_realm"."restricted_to_domain", "zephyr_recipient"."id", "zephyr_recipient"."type_id", "zephyr_recipient"."type", "zephyr_client"."id", "zephyr_client"."name" FROM "zephyr_message" INNER JOIN "zephyr_userprofile" ON ( "zephyr_message"."sender_id" = "zephyr_userprofile"."id" ) INNER JOIN "zephyr_realm" ON ( "zephyr_userprofile"."realm_id" = "zephyr_realm"."id" ) INNER JOIN "zephyr_recipient" ON ( "zephyr_message"."recipient_id" = "zephyr_recipient"."id" ) INNER JOIN "zephyr_client" ON ( "zephyr_message"."sending_client_id" = "zephyr_client"."id" ) ORDER BY "zephyr_message"."id" DESC LIMIT 25000;
with query plan:
Limit (cost=0.00..27120.95 rows=25000 width=362) (actual time=0.051..1121.282 rows=25000 loops=1)
-> Nested Loop (cost=0.00..5330872.99 rows=4913981 width=362) (actual time=0.048..1081.014 rows=25000 loops=1)
-> Nested Loop (cost=0.00..3932643.31 rows=4913981 width=344) (actual time=0.042..926.398 rows=25000 loops=1)
-> Nested Loop (cost=0.00..2550275.29 rows=4913981 width=334) (actual time=0.035..752.524 rows=25000 loops=1)
Join Filter: (zephyr_message.sending_client_id = zephyr_client.id)
-> Nested Loop (cost=0.00..1739467.29 rows=4913981 width=320) (actual time=0.024..217.348 rows=25000 loops=1)
-> Index Scan Backward using zephyr_message_pkey on zephyr_message (cost=0.00..362510.09 rows=4913981 width=156) (actual time=0.014..42.097 rows=25000 loops=1)
-> Index Scan using zephyr_userprofile_pkey on zephyr_userprofile (cost=0.00..0.27 rows=1 width=164) (actual time=0.003..0.004 rows=1 loops=25000)
Index Cond: (id = zephyr_message.sender_id)
-> Materialize (cost=0.00..1.17 rows=11 width=14) (actual time=0.001..0.010 rows=11 loops=25000)
-> Seq Scan on zephyr_client (cost=0.00..1.11 rows=11 width=14) (actual time=0.002..0.010 rows=11 loops=1)
-> Index Scan using zephyr_recipient_pkey on zephyr_recipient (cost=0.00..0.27 rows=1 width=10) (actual time=0.002..0.003 rows=1 loops=25000)
Index Cond: (id = zephyr_message.recipient_id)
-> Index Scan using zephyr_realm_pkey on zephyr_realm (cost=0.00..0.27 rows=1 width=18) (actual time=0.002..0.003 rows=1 loops=25000)
Index Cond: (id = zephyr_userprofile.realm_id)
Total runtime: 1141.408 ms
In the new code, we do two queries:
SELECT "zephyr_message"."id" FROM "zephyr_message" ORDER BY "zephyr_message"."id" DESC LIMIT 1
followed by:
SELECT "zephyr_message"."id", "zephyr_message"."sender_id", "zephyr_message"."recipient_id", "zephyr_message"."subject", "zephyr_message"."content", "zephyr_message"."rendered_content", "zephyr_message"."rendered_content_version", "zephyr_message"."pub_date", "zephyr_message"."sending_client_id", "zephyr_userprofile"."id", "zephyr_userprofile"."password", "zephyr_userprofile"."last_login", "zephyr_userprofile"."email", "zephyr_userprofile"."is_staff", "zephyr_userprofile"."is_active", "zephyr_userprofile"."date_joined", "zephyr_userprofile"."full_name", "zephyr_userprofile"."short_name", "zephyr_userprofile"."pointer", "zephyr_userprofile"."last_pointer_updater", "zephyr_userprofile"."realm_id", "zephyr_userprofile"."api_key", "zephyr_userprofile"."enable_desktop_notifications", "zephyr_userprofile"."enter_sends", "zephyr_userprofile"."tutorial_status", "zephyr_realm"."id", "zephyr_realm"."domain", "zephyr_realm"."restricted_to_domain", "zephyr_recipient"."id", "zephyr_recipient"."type_id", "zephyr_recipient"."type", "zephyr_client"."id", "zephyr_client"."name" FROM "zephyr_message" INNER JOIN "zephyr_userprofile" ON ( "zephyr_message"."sender_id" = "zephyr_userprofile"."id" ) INNER JOIN "zephyr_realm" ON ( "zephyr_userprofile"."realm_id" = "zephyr_realm"."id" ) INNER JOIN "zephyr_recipient" ON ( "zephyr_message"."recipient_id" = "zephyr_recipient"."id" ) INNER JOIN "zephyr_client" ON ( "zephyr_message"."sending_client_id" = "zephyr_client"."id" ) WHERE "zephyr_message"."id" > 4941883
with the message id filled in as the result of the first query. The
new query differs from the original only in that its ORDER BY and
LIMIT clauses are replaced by a WHERE clause. The second query has
query plan:
Hash Join (cost=709.30..28048.18 rows=20544 width=365) (actual time=41.678..279.261 rows=25041 loops=1)
Hash Cond: (zephyr_message.recipient_id = zephyr_recipient.id)
-> Hash Join (cost=102.98..27056.66 rows=20544 width=355) (actual time=3.686..190.730 rows=25041 loops=1)
Hash Cond: (zephyr_message.sending_client_id = zephyr_client.id)
-> Hash Join (cost=101.73..26772.94 rows=20544 width=341) (actual time=3.649..143.695 rows=25041 loops=1)
Hash Cond: (zephyr_userprofile.realm_id = zephyr_realm.id)
-> Hash Join (cost=99.99..26488.71 rows=20544 width=323) (actual time=3.578..96.746 rows=25041 loops=1)
Hash Cond: (zephyr_message.sender_id = zephyr_userprofile.id)
-> Index Scan using zephyr_message_pkey on zephyr_message (cost=0.00..26106.24 rows=20544 width=159) (actual time=0.017..41.980 rows=25041 loops=1)
Index Cond: (id > 4941883)
-> Hash (cost=83.33..83.33 rows=1333 width=164) (actual time=3.548..3.548 rows=1333 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 275kB
-> Seq Scan on zephyr_userprofile (cost=0.00..83.33 rows=1333 width=164) (actual time=0.006..1.646 rows=1333 loops=1)
-> Hash (cost=1.33..1.33 rows=33 width=18) (actual time=0.064..0.064 rows=33 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 2kB
-> Seq Scan on zephyr_realm (cost=0.00..1.33 rows=33 width=18) (actual time=0.003..0.033 rows=33 loops=1)
-> Hash (cost=1.11..1.11 rows=11 width=14) (actual time=0.027..0.027 rows=11 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 1kB
-> Seq Scan on zephyr_client (cost=0.00..1.11 rows=11 width=14) (actual time=0.003..0.013 rows=11 loops=1)
-> Hash (cost=335.03..335.03 rows=21703 width=10) (actual time=37.974..37.974 rows=21761 loops=1)
Buckets: 4096 Batches: 1 Memory Usage: 893kB
-> Seq Scan on zephyr_recipient (cost=0.00..335.03 rows=21703 width=10) (actual time=0.004..18.443 rows=21761 loops=1)
Total runtime: 299.300 ms
(imported from commit b2a70cccc47be7970df407c6be00eccd2e8be82a)
When you create a stream that you'd previously created (then unsubscribed from),
it was possible to end up in the subscribers list twice. Once came from loading
the subscribers list from the backend, and once came from a bit of mark_subscribed
logic that only gets called if you've subscribed to that stream at least once before
in the current session.
resolves trac #1196
(imported from commit e47ff139a9c25b1b8689ea6795dfad96ae8d2591)
If the pasted content has strings, we don't upload included files and instead
allow the default behavior to take place. This deals with a quirky behavior of
pastes from MS Word, which in addition to the formatted string content also
includes a thumbnail of it. Images still paste as usual.
(imported from commit 60c4f8dd90ac2e8e38940fb302cc9d1ebeecfdf3)
This allows users on signup-eligible domains to sign up for Humbug using
Google Apps.
As part of this, we wrap the openid done view in our own code in order to
handle the "Unknown user" error. Therein, we create a PreregistrationUser
and then shunt the user through the rest of the confirmation process, pre-
filling in their name.
(imported from commit 066d9a1021384a6da2662352e62a701451bd6f44)
Changes include:
* New markup for the button in compose.html
* A hidden file input field in compose.html
* Added reference to the file input field in filedrop
initialization in compose.js
* A feature test and a click event binding for
the "Attach files" button in ui.js
* New paperclip icon reference in fonts.css
* New general hidden display classes in zephyr.css
* New composition pane button classes in zephyr.css
Fixes to the "Attach files" button commit e673bda...
Changes include:
* Fixed the feature test for (new XMLHttpRequest).upload so
it works in Firefox.
* Renamed .button to .message-control-button
* Removed stray newlines
(imported from commit c1f0834b74fd7120ec27db64ec380ffb3fa34633)
Having a message ID range significantly improves the query
performance because the number of messages Postgres has to consider
is much smaller.
(imported from commit 9b007457712f1c1502d526abea1b6fd742bd911d)
The fact that we were dumping this cache and not refilling it seems to
be one of the causes of Tornado restarts being a lot slower on prod
than on local systems.
(imported from commit a32a759f4dfb591706ede1cce2d38f5c3704193c)
Previously, our check for whether we needed to call load_old_messages
a second time on page load to get up to the present caused us to
basically always do such a call.
(imported from commit b599041e8c0853b4c8c9ab2def6679142302523e)
On my laptop, this saves about 80 milliseconds per 1000 messages
requested via get_old_messages queries. Since we only have one
memcached process and it does not run with special priority, this
might have significant impact on load during server restarts.
(imported from commit 06ad13f32f4a6d87a0664c96297ef9843f410ac5)
The internal format of 'message' had changed, so prior to this commit,
the tutorial was receiving (a) internally inconsistent, and (b)
not-what-it-expected versions of the message.
(imported from commit 233b934e6b600bd59125d133fdf7443fd8f6bbf8)
It's subtle, but the slice was in the wrong place and wasn't
actually truncating the stream name at all, so the client and
server disagreed about where the tutorial messages should go.
(It might be the case that we should accept the tutorial stream
name from the client directly, rather than computing it in two
places.)
(imported from commit 8273223f182e8ad36eaea1cbf75e1426fcfdfbab)
If the system was waiting for you to reply and you replied 'exit', the
tutorial would stop -- but our thing that was waiting for you to reply
would continue waiting. It would eventually timeout and send you the
heartbroken "I didn't hear from you so I stopped waiting" message.
Chances are, you were unsubscribed so you didn't see it, but we
should still just not send it.
(imported from commit 694e442bc29b32efd59f08b4b8b5f573768aea21)
Previously it was centered with respect to its enclosing div, which
looked slightly off.
(imported from commit 3878f162d3eb50ce85cae7054102095069aa60c8)
Pretty hackish for now since this is presumably going to all
be redone with Font Awesome icons in not too long.
(imported from commit 497d6cf18d7a8d6014a20c08d66d88c324478e55)
Timing out within the Twitter portion of the render causes the message
to still go through (without a preview). If we don't timeout here, it
causes the entire Markdown render to timeout, which rejects the
message in its entirety -- a far worse outcome.
(imported from commit f510a56f48afa46da8ec6277496fa03374cdb042)
This was apparently broken by the final revision of our fix to the
autoscrolling+narrow bugs, because it attempted to use jquery's
animation queues to restrict which animations were stopped, and this
doesn't seem to work.
(imported from commit cf97f9f56dc5a16d1aa0322b5e6ec432a76d3be2)
See PEP 328[1] for details. This feature was introduced in Python 2.5 and
will become mandatory in Python 3.
[1]: http://www.python.org/dev/peps/pep-0328
(imported from commit 7444eeba8a08d5f91b94c7921848f2274979bd76)
Don't assume clipboardData.items since it doesn't exist on Safari
Make sure there are no files if using a clipboard drop. Safari includes a blank text/uri-list
data entry
Firefox fix for image pasting
(imported from commit ea0d56fe73ca45cf2e4d437df23a4023bb649445)
Previously, we were calling util.same_stream_and_subject on a pair of
messages, one of which was a private message, which is not valid. We
should have instead been calling util.same_recipient, which checks the
message type as well.
(imported from commit bc5715807036bff1fd4f214dafad00e33678e91d)
Leo Franchi
acrefoot
Waseem Daher
Allen Rabinovich
Zev Benjamin
Tim Abbott
Steve Howell
Jessica McKellar
Luke Faraone
Jeff Arnold