Turn on `SO_LINGER` for client communication sockets. #30

pcwalton · 2016-01-22T20:07:13Z

This reduces the probability that the receiver receives errors when we
close our end of the socket with data remaining.

There is deadlock potential with this patch, because turning on SO_LINGER
causes close() to block until the receiver has received all the data. If
deadlocks happen, a workaround will be to close sockets in a separate thread.
This is ugly and slow, so I don't want to do that unless we need to.

Might fix #29.

r? @jdm

jdm · 2016-01-22T20:13:18Z

This means that the socket hangs around for 30s before its data becomes unavailable like before, right?

pcwalton · 2016-01-22T20:56:05Z

It's 30 seconds or whenever the receiver has read all the data, whichever comes first. The 30 second timeout should hopefully never happen in practice--it would only happen if the receiver leaked its receiving end for some reason.

jdm · 2016-01-22T21:02:42Z

Well, it's not hard for me to imagine cases where the receiver is not sitting waiting synchronously and is occasionally polling (the equivalent of setTimeout, basically), but this seems like it should be an improvement over the status quo.

jdm · 2016-01-22T21:03:15Z

@bors-servo: r+

bors-servo · 2016-01-22T21:03:17Z

📌 Commit 153f474 has been approved by jdm

bors-servo · 2016-01-22T21:03:18Z

⌛ Testing commit 153f474 with merge db5280a...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-22T21:14:02Z

💔 Test failed - travis

jdm · 2016-01-22T21:19:21Z

@bors-servo: retry

bors-servo · 2016-01-22T21:19:23Z

⌛ Testing commit 153f474 with merge 619a420...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-22T21:29:58Z

💔 Test failed - travis

jdm · 2016-01-22T21:33:50Z

@bors-servo: retry

bors-servo · 2016-01-22T21:33:52Z

⌛ Testing commit 153f474 with merge 28dce6a...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-22T21:41:17Z

💔 Test failed - travis

jdm · 2016-01-22T21:54:12Z

@bors-servo: retry

bors-servo · 2016-01-22T21:54:13Z

⌛ Testing commit 153f474 with merge ea0eb7e...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-22T22:02:28Z

💔 Test failed - travis

jdm · 2016-01-23T16:07:43Z

@bors-servo: retry

bors-servo · 2016-01-23T16:07:44Z

⌛ Testing commit 153f474 with merge b2bd35f...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-23T16:09:12Z

💔 Test failed - travis

jdm · 2016-01-23T16:17:29Z

Will need to wait for either https://github.com/Manishearth/rust-clippy/issues/576 or serde-deprecated/aster#63.

jdm · 2016-01-25T03:55:10Z

@bors-servo: retry

bors-servo · 2016-01-25T03:55:11Z

⌛ Testing commit 153f474 with merge c5bec7a...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-25T03:56:56Z

💔 Test failed - travis

jdm · 2016-01-25T04:00:15Z

platform/linux/mod.rs:499:18: 499:24 error: `linger` does not name a structure [E0422]

platform/linux/mod.rs:499     let linger = linger {

                                           ^~~~~~

platform/linux/mod.rs:499:18: 499:24 help: run `rustc --explain E0422` to see a detailed explanation

platform/linux/mod.rs:504:32: 504:42 error: unresolved name `SOL_SOCKET` [E0425]

platform/linux/mod.rs:504                                SOL_SOCKET,

                                                         ^~~~~~~~~~

platform/linux/mod.rs:504:32: 504:42 help: run `rustc --explain E0425` to see a detailed explanation

platform/linux/mod.rs:506:50: 506:56 error: use of undeclared type name `linger` [E0412]

platform/linux/mod.rs:506                                &linger as *const linger as *const c_void,

                                                                           ^~~~~~

platform/linux/mod.rs:506:50: 506:56 help: run `rustc --explain E0412` to see a detailed explanation

platform/linux/mod.rs:507:47: 507:53 error: use of undeclared type name `linger` [E0412]

platform/linux/mod.rs:507                                mem::size_of::<linger>() as socklen_t);

                                                                        ^~~~~~

platform/linux/mod.rs:507:47: 507:53 help: run `rustc --explain E0412` to see a detailed explanation

platform/linux/mod.rs:499:18: 499:24 error: `linger` does not name a structure [E0422]

platform/linux/mod.rs:499     let linger = linger {

                                           ^~~~~~

platform/linux/mod.rs:499:18: 499:24 help: run `rustc --explain E0422` to see a detailed explanation

platform/linux/mod.rs:504:32: 504:42 error: unresolved name `SOL_SOCKET` [E0425]

platform/linux/mod.rs:504                                SOL_SOCKET,

                                                         ^~~~~~~~~~

platform/linux/mod.rs:504:32: 504:42 help: run `rustc --explain E0425` to see a detailed explanation

platform/linux/mod.rs:506:50: 506:56 error: use of undeclared type name `linger` [E0412]

platform/linux/mod.rs:506                                &linger as *const linger as *const c_void,

                                                                           ^~~~~~

platform/linux/mod.rs:506:50: 506:56 help: run `rustc --explain E0412` to see a detailed explanation

platform/linux/mod.rs:507:47: 507:53 error: use of undeclared type name `linger` [E0412]

platform/linux/mod.rs:507                                mem::size_of::<linger>() as socklen_t);

                                                                        ^~~~~~

Incomplete changes?


        Turn on `SO_LINGER` for client communication sockets.

This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29.

pcwalton · 2016-01-25T18:47:08Z

Forgot to commit changes. r? @jdm

jdm · 2016-01-25T18:49:21Z

@bors-servo: r+

bors-servo · 2016-01-25T18:49:22Z

📌 Commit 0146ae7 has been approved by jdm

bors-servo · 2016-01-25T18:49:23Z

⌛ Testing commit 0146ae7 with merge 3a95bec...


        Auto merge of #30 - pcwalton:linger, r=jdm

@jdm

Turn on `SO_LINGER` for client communication sockets. This reduces the probability that the receiver receives errors when we close our end of the socket with data remaining. There is deadlock potential with this patch, because turning on `SO_LINGER` causes `close()` to block until the receiver has received all the data. If deadlocks happen, a workaround will be to close sockets in a separate thread. This is ugly and slow, so I don't want to do that unless we need to. Might fix #29. r? @jdm

bors-servo · 2016-01-25T18:59:34Z

☀️ Test successful - travis


        Test that data is recvable after sender is dropped

Add test for servo#30. Test that data can be retrieved from the receiver after the sender has been dropped.


        Test: data is recv-able after sender is dropped

Add test for servo#30. Test that data can be retrieved from the receiver after the sender has been dropped.


        Auto merge of #58 - danlrobertson:i30, r=jdm

Test: data is recv-able after sender is dropped Add test for #30. Test that data can be retrieved from the receiver after the sender has been dropped. Fixes: #33  --- This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/ipc-channel/58)

jdm mentioned this pull request Jan 25, 2016

Intermittent Result::unwrap() panic in WPT reftest ("Connection reset by peer" or "Couldn't receive FontCacheThread") servo/servo#8815

Closed

pcwalton force-pushed the pcwalton:linger branch from 153f474 to 0146ae7 Jan 25, 2016

bors-servo merged commit 0146ae7 into servo:master Jan 25, 2016
1 check was pending

1 check was pending

homu Testing commit 0146ae7 with merge 3a95bec...
Details

dlrobertson added a commit to dlrobertson/ipc-channel that referenced this pull request Mar 28, 2016

Test that data is recvable after sender is dropped

a97a2b2

Add test for servo#30. Test that data can be retrieved from the receiver after the sender has been dropped.

dlrobertson added a commit to dlrobertson/ipc-channel that referenced this pull request Mar 28, 2016

Test: data is recv-able after sender is dropped

851c64d

Add test for servo#30. Test that data can be retrieved from the receiver after the sender has been dropped.

dlrobertson mentioned this pull request Mar 28, 2016

Test: data is recv-able after sender is dropped #58

Merged

servo / ipc-channel

Turn on `SO_LINGER` for client communication sockets. #30

Turn on `SO_LINGER` for client communication sockets. #30

pcwalton commented Jan 22, 2016

jdm commented Jan 22, 2016

pcwalton commented Jan 22, 2016

jdm commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 23, 2016

bors-servo commented Jan 23, 2016

bors-servo commented Jan 23, 2016

jdm commented Jan 23, 2016

jdm commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016

jdm commented Jan 25, 2016

pcwalton commented Jan 25, 2016

jdm commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016

servo / ipc-channel

Join GitHub today

Turn on `SO_LINGER` for client communication sockets. #30

Turn on `SO_LINGER` for client communication sockets. #30

Conversation

pcwalton commented Jan 22, 2016

jdm commented Jan 22, 2016

pcwalton commented Jan 22, 2016

jdm commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 22, 2016

bors-servo commented Jan 22, 2016

bors-servo commented Jan 22, 2016

jdm commented Jan 23, 2016

bors-servo commented Jan 23, 2016

bors-servo commented Jan 23, 2016

jdm commented Jan 23, 2016

jdm commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016

jdm commented Jan 25, 2016

pcwalton commented Jan 25, 2016

jdm commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016

bors-servo commented Jan 25, 2016