Summary:
Ref T5373. The control flow between `aphlict-listener` and `JX.Aphlict` is pretty weird right now, where the listener (which is the highest-level component) has intimate knowledge of how to put the SWF on the page.
Instead:
- Make `JX.Aphlict` a real singleton.
- Instantiate it sooner.
- Have it handle the flash setup handshake.
Test Plan: Loaded page in debug mode, saw normal flow take place.
Reviewers: joshuaspence
Reviewed By: joshuaspence
Subscribers: epriestley
Maniphest Tasks: T5373
Differential Revision: https://secure.phabricator.com/D9699
Summary:
Ref T5373. This seems to work pretty much correctly.
Also stop popping bubbles and just use the log, since users find the bubbles confusing/not useful and they're not great for developers either.
Future diffs will expose more user-facing stuff.
Test Plan: Added `throw` to AphlictClient.as, got a log in the parent window.
Reviewers: joshuaspence
Reviewed By: joshuaspence
Subscribers: epriestley
Maniphest Tasks: T5373
Differential Revision: https://secure.phabricator.com/D9698
Summary: I don't believe that this HTML serves a purpose any longer. In addition, loading https://secure.phabricator.com/rsrc/swf/test.html doesn't seem particularly useful.
Test Plan: N/A
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9691
Summary: Ref T5284. When an `AphlictClient` is closed, it will eventually be purged from the pool by the `AphlictMaster`. When this happens, also unsubscribe the purged client from all notifications, and send an `unsubscribe` command to the Aphlict server if possible.
Test Plan:
Verified the output of the Aphlict server (running in debug mode):
```
[Wed Jun 11 2014 23:21:31 GMT+0000 (UTC)] <FlashPolicy> Policy Request From ::ffff:192.168.1.1
[Wed Jun 11 2014 23:21:31 GMT+0000 (UTC)] <Listener/2> Connected from ::ffff:192.168.1.1
[Wed Jun 11 2014 23:21:31 GMT+0000 (UTC)] <Listener/2> Received data: {"command":"subscribe","data":["PHID-USER-cb5af6p4oepy5tlgqypi"]}
[Wed Jun 11 2014 23:21:31 GMT+0000 (UTC)] <Listener/2> Subscribed to: ["PHID-USER-cb5af6p4oepy5tlgqypi"]
[Wed Jun 11 2014 23:21:39 GMT+0000 (UTC)] <Listener/2> Received data: {"command":"subscribe","data":["PHID-TASK-l2dtbs5xrt2b7abgh5a6"]}
[Wed Jun 11 2014 23:21:39 GMT+0000 (UTC)] <Listener/2> Subscribed to: ["PHID-TASK-l2dtbs5xrt2b7abgh5a6"]
[Wed Jun 11 2014 23:21:57 GMT+0000 (UTC)] <Listener/2> Received data: {"command":"unsubscribe","data":["PHID-TASK-l2dtbs5xrt2b7abgh5a6"]}
[Wed Jun 11 2014 23:21:57 GMT+0000 (UTC)] <Listener/2> Unsubscribed from: ["PHID-TASK-l2dtbs5xrt2b7abgh5a6"]
```
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Maniphest Tasks: T5284
Differential Revision: https://secure.phabricator.com/D9492
Summary: Currently, the `AphlictClient` will only send its subscriptions to the `AphlictMaster` once. If the original `AphlictMaster` is closed and a new master is created, then client subscriptions will be lost.
Test Plan: Opened two separate tabs. Closed the "master" tab and noticed that the subscriptions were re-sent to the server.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9487
Summary: Ref T4324. Ref T5284. This adds server-side support for keeping track of a set of PHIDs that the Aphlict clients have subscribed to. Instead of broadcasting a notification to all clients (after which the clients can poll `/notification/individual` in order to determine whether or not they are interested in the notification), transmit notifications only to clients that have subscribed to a PHID that is relevant to the notification.
Test Plan:
I opened up two clients on the same host (incognito tabs in Chrome). Here is the output from the server:
```
> sudo ./bin/aphlict debug
Starting Aphlict server in foreground...
Launching server:
$ 'nodejs' '/usr/src/phabricator/src/applications/aphlict/management/../../../../support/aphlict/server/aphlict_server.js' --port='22280' --admin='22281' --host='localhost' --user='aphlict'
[Wed Jun 11 2014 19:10:27 GMT+0000 (UTC)] Started Server (PID 4546)
[Wed Jun 11 2014 19:10:36 GMT+0000 (UTC)] <FlashPolicy> Policy Request From ::ffff:192.168.1.1
[Wed Jun 11 2014 19:10:37 GMT+0000 (UTC)] <Listener/1> Connected from ::ffff:192.168.1.1
[Wed Jun 11 2014 19:10:37 GMT+0000 (UTC)] <Listener/1> Received data: {"command":"subscribe","data":["PHID-USER-cb5af6p4oepy5tlgqypi"]}
[Wed Jun 11 2014 19:10:37 GMT+0000 (UTC)] <Listener/1> Subscribed to: ["PHID-USER-cb5af6p4oepy5tlgqypi"]
[Wed Jun 11 2014 19:10:39 GMT+0000 (UTC)] <Listener/1> Received data: {"command":"subscribe","data":["PHID-USER-kfohe3ca5oe6ygykmioq"]}
[Wed Jun 11 2014 19:10:39 GMT+0000 (UTC)] <Listener/1> Subscribed to: ["PHID-USER-kfohe3ca5oe6ygykmioq"]
[Wed Jun 11 2014 19:10:42 GMT+0000 (UTC)] notification: {"key":"6023751084283587681","type":"notification","subscribers":["PHID-USER-cb5af6p4oepy5tlgqypi"]}
[Wed Jun 11 2014 19:10:42 GMT+0000 (UTC)] <Listener/1> Wrote Message
```
I verified (using the "Network" tab in Chrome) that an AJAX request to `/notification/individual/` was only made in the tab belonging to the user that triggered the test notification.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Maniphest Tasks: T5284, T4324
Differential Revision: https://secure.phabricator.com/D9458
Summary:
Currently, the Aphlict server will crash if invalid JSON data is `POST`ed to it. I have fixed this to, instead, return a 400. Also made some minor formatting changes.
Ref T4324. Ref T5284. Also, modify the data structure that is passed around (i.e. `POST`ed to the Aphlict server and broadcast to the Aphlict clients) to include the subscribers. Initially, I figured that we shouldn't expose this information to the clients... however, it is necessary for T4324 that the `AphlictMaster` is able to route a notification to the appropriate clients.
Test Plan:
Making the following `curl` request: `curl --data "{" http://localhost:22281/`.
**Before**
```
sudo ./bin/aphlict debug
Starting Aphlict server in foreground...
Launching server:
$ 'nodejs' '/usr/src/phabricator/src/applications/aphlict/management/../../../../support/aphlict/server/aphlict_server.js' --port='22280' --admin='22281' --host='localhost' --user='aphlict'
[Wed Jun 11 2014 17:07:51 GMT+0000 (UTC)] Started Server (PID 2033)
[Wed Jun 11 2014 17:07:55 GMT+0000 (UTC)]
<<< UNCAUGHT EXCEPTION! >>>
SyntaxError: Unexpected end of input
>>> Server exited!
```
**After**
(No output... the bad JSON is caught and a 400 is returned)
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Maniphest Tasks: T4324, T5284
Differential Revision: https://secure.phabricator.com/D9480
Summary: Currently, any error thrown when instantiating an `AphlictMaster` will be assumed to be due to the master already existing. This is a bit overzealous because the [[http://help.adobe.com/en_US/FlashPlatform/reference/actionscript/3/flash/net/LocalConnection.html#connect() | documentation]] specifically states than an `ArgumentError` will be throw if "the `LocalConnection` instance is already connected".
Test Plan: Inspected the log message.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9422
Summary: These methods shouldn't be overridden. Marking them as `final` will enforce this.
Test Plan: N/A
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9411
Summary: Ref T4324. Currently, if the `AphlictMaster` receives dodgy data from the Aphlict server (invalid JSON, for example) then a syntax error will be thrown and the `AphlictMaster` will die. Instead, catch errors and raise a notification.
Test Plan: {F163466}
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Maniphest Tasks: T4324
Differential Revision: https://secure.phabricator.com/D9380
Summary:
Ref T4324. Currently, an Aphlict client (with a corresponding connection to the Aphlict Server) is created for every tab that a user has open. This significantly affects the scalability of Aphlict as a service. Instead, we can use `LocalConnection` instances to coordinate the communication of multiple Aphlict clients to the server.
Similar functionality existed prior to D2704, but was removed as the author was not able to get this functionality working as intended. It seems that the main issue with the initial attempt was the use of the `setTimeout` function, which seemed to be a blocking call which prevented messages from being received. I have instead used an event-based model using a `Timer` object.
Roughly this works as follows:
# The first instance will create an `AphlictClient` and an `AphlictMaster`. The `AphlictClient` will register itself with the `AphlictMaster` and will consequently be notified of incoming messages.
# The `AphlictClient` is then responsible for pinging the `AphlictMaster` at regular intervals. If the client does not ping the master in a given period of time, the master will assume that the client is dead and will remove the client from the pool.
# Similarly, the `AphlictMaster` is required to respond to pings with a "pong" response. The pong response lets the clients know that the `AphlictMaster` is still alive. If the clients do not receive a pong in a given period of time, then the clients will attempt to spawn a new master.
Test Plan: I have tested this on our Phabricator install with a few tabs opened and inspecting the console output. I will upload a screencast of my test results.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Maniphest Tasks: T4324
Differential Revision: https://secure.phabricator.com/D9327
Summary:
Adds the node.js Aphlict server, the flash Aphlict client, and some
supporting javascript. Built on top of - and requires - D2703 (which is still
in progress). Will likely work with no modification on top of the final
version, though.
The node server is currently run with
sudo node support/aphlict/server/aphlict_server.js
Test Plan: tested locally
Reviewers: epriestley
Reviewed By: epriestley
CC: allenjohnashton, keebuhm, aran, Korvin
Differential Revision: https://secure.phabricator.com/D2704
Summary: Provide a reasonable JS API for the Aphlict client. Provide an example behavior to invoke it.
Test Plan:
Ran "aphlict_server.js" with:
$ sudo node aphlict_server.js
Loaded /aphlict/. Opened console. Got "hello" from the server every second.
Got reasonable errors with the server not present ("Security exception", but this is because it can't connect to port 843 to access the policy server).
Reviewers: ddfisher, keebuhm, allenjohnashton, btrahan
Reviewed By: btrahan
CC: aran, epriestley
Maniphest Tasks: T944
Differential Revision: https://secure.phabricator.com/D1800
Summary:
This is purely a prototype at the moment, but the basic functionality sort of
works.
I'm not sure how far I want to go with this but I think we might be able to get
somewhere without it being gross.
The idea here is to build a notification server WITHOUT using Comet, since Comet
is extremely difficult and complicated.
Instead, I use Flash on the client. LocalConnection allows flash instances to
talk to each other and connect() can be used as a locking primitive. This allows
all the instances to elect a master instance in a race-safe way. The master is
responsible for opening a single connnection to the server.
On the server, I use Node.js since PHP is pretty unsuitable for this task.
See Github Issue #3: https://github.com/facebook/phabricator/issues/3
One thing I need to figure out next is if I can reasonably do SSL/TSL over Flash
(it looks like I can, in theory, with the as3crypto library) or if the server
needs to just send down version information and trigger a separate Ajax call on
the client.
Test Plan:
Created a client pool and connected it to the server, with election and failover
apparently working correctly.
Reviewed By: aran
Reviewers: Girish, aran, jungejason, tuomaspelkonen, davidrecordon
Commenters: Girish, davidrecordon
CC: aran, epriestley, Girish, davidrecordon
Differential Revision: 284
