Tab discarder: fine tune the killing rate |
||||||||||||||
Issue descriptionCurrently when receiving a critical memory pressure, tab discarder calculates "target memory to free" using low memory margin, available memory, and estimation of memory freed by killing each candidate. However 1. Current estimation uses RSS of the renderer process when killing a tab, which is inaccurate because of a renderer process might host multiple tabs. 2. A minor problem: in the process of killing candidate, memory level has been changed because of swap of other factors. To be more accurate, we can check available memory before each kill. I have a CL which polls /dev/chromeos-low-mem before each kill to decide whether to continue: https://chromium-review.googlesource.com/c/523043 However as yusukes pointed out, it'll not work well because app kills are not synchronous, tab kills are potentially not either. So memory would not be freed immediately. It'll be better if we can find a better way to control the killing rate. For example, is it possible to setup a second tier polling inside tab discarder? Like when receiving a critical memory pressure, poll /dev/chromeos-low-mem 5 times a second?
,
Dec 1 2017
,
Dec 1 2017
Since there may be activity on this bug, I'd like to add some information, which leads me as well to speculate that polling 5x per second is not going to help. In the attached graph, blue vertical lines are the times when the kernel signals a low-memory state. That happens at the crossing of the available memory (blue graph) and the margin (blue horizontal line). Red lines represent tab discards. This is a bit of an extreme situation (lots of open tabs, lots of tab switchings and reloads) but as you can see the tab discards go on for about 40 seconds after the last notification. Also one of the tabs was very large and it either crashed or was discarded, so at t = 12 seconds there is no longer a need to discard anything, yet 20+ more tabs are discarded.
,
Dec 1 2017
Luigi, you know way more about this than I do, so passing it over to you :) If you feel we should just WontFix this, feel free to.
,
Dec 1 2017
Well how kind of you. That teaches me to keep my mouth shut. ;) I mostly work on the kernel, and as we speak I am writing a tool to collect fine-grained memory stats. That's why I have this information. But I've hacked Chrome too and have no excuse for pushing back on this :/ I don't know what you're working on or what interests you, but if you like this topic, don't be intimidated, I'll be happy to collaborate.
,
Dec 2 2017
Please see #6 :)
,
Dec 2 2017
,
Dec 4 2017
Hi Luigi, I'm kind of confused. The bug refers to the problem where when one notification comes in, the "for" loop in tab discarder would 1) Get a snapshot of current memory status and 2) Kill tabs until memory backs to normal level according to the info got from 1) and estimated memory to be freed for each tab. The whole "for" loop after getting a notification should be executed in under 1 second. How could a notification causes a tab to be discarded after ~40 seconds? It looks very strange to me. Do you have any assumption ? A new bug ?
,
Dec 4 2017
Hi Cheng-Yu, yes I agree it looks strange, but I am not familiar with that code so I don't have an explanation. I just collected the data and put it on the same graph. If we don't trust the data, we may be able to confirm these events by looking at logs. Bear in mind that the system was under very heavy load, with many tabs reloading. The data is collected in a separate daemon, and Chrome notifies the daemon via DBus. I added the tab discard signal here: https://chromium-review.googlesource.com/c/chromium/src/+/754301/2/chrome/browser/resource_coordinator/tab_manager.cc Could that be the problem?
,
Jan 10 2018
,
Jan 12 2018
Hi Luigi,
Per #3, how did you get the timing of kernel signals a low-memory state (vertical blue line)? The code you showed in #10 is actually not when tab discarder happens. Instead it should be called every time when system is in low memory state. Memory coordinator polls the kernel interface for available memory periodically (like 1 second) and check whether there's room for several renderers. The new logic is like
int expected_renderer_count = available / kExpectedRendererSizeMB;
if (available <= 0 || expected_renderer_count < kNewRenderersUntilCritical)
return MemoryCondition::CRITICAL;
https://cs.chromium.org/chromium/src/content/browser/memory/memory_condition_observer.cc?rcl=4bccc9372c2644679beecf3e3a88b3630ec32132&l=78
I'm not sure whether memory coordinator has been enabled in the Chrome version you used. But the old code did roughly the same. I have no idea why it continues to fire when there's lots of available memory.
On the other hand, when PurgeMemoryAndDiscardTab is called, on ChromeOS we have another inner loop which reads the low memory margin and tries to kill several tabs to reclaim enough memory
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager_delegate_chromeos.cc?rcl=5b0879c77d86676a28d6e4334c3bdca0f005f6cb&l=650
If when this function is called the low memory margin is negative (enough has been reclaimed), no tab would be discarded at all. If you want to know when a tab is really discarded, you should add a log at KillTab() and KillArcProcess() in this file.
,
Jan 12 2018
The low-memory crossing (vertical blue line) is directly from /dev/chromeos-low-mem. But I see what you're saying, manager->DiscardTab doesn't necessarily discard a tab. Although the comment in DiscardTabImpl says: // Loop until a non-discarded tab to kill is found. But you're right, if the kArcMemoryManagement feature is enabled then it calls LowMemoryKillImpl, which does not necessarily free tabs. In any case, it is strange that it is called so many times when there seems to be a lot of free RAM. Maybe the low-memory condition is computed differently? I am not sure if these calls are expensive, but they pollute the logs. Should I file a separate bug for this?
,
Jan 13 2018
Checked the code again, MemoryCoordinator defaults to disabled so the only place where critical memory pressure occurs looks still https://cs.chromium.org/chromium/src/base/memory/memory_pressure_monitor_chromeos.cc?rcl=7a7c695b9e11047c95c371446f0939ece09c8ac3&l=199 where /dev/chromeos-low-mem is set. Does the value "available' come from /sys/kernel/mm/chromeos-low_mem/available ? And it should be the only code path where PurgeMemoryAndDiscardTab() is called (when critical memory pressure happens) https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=7a7c695b9e11047c95c371446f0939ece09c8ac3&l=887 Have no idea why the call is not synced with /dev/chromeos-low-mem I tried to log available memory on my cave when PurMemoryAndDiscardTab() is called and it's close to low memory margin (< ±30MB). Can you reliably reproduce the problem? I would like to know the steps so I can look into this. It's really weird. Is there any possibility that the red lines don't align with blue lines properly?
,
Jan 16 2018
#14 hi! Sorry, I don't follow your reasoning on the code. But to answer your questions: > Does the value "available' come from /sys/kernel/mm/chromeos-low_mem/available ? I don't know. In theory it should, but there is code to calculate it in https://cs.chromium.org/chromium/src/base/memory/memory_pressure_monitor_chromeos.cc I hope it's not used because it doesn't match what we do on caroline. > Can you reliably reproduce the problem? I would like to know the steps so I can look into this. I load a large number of CNN tabs and Google properties, multiple times, for a total of 30 to 50 tabs, and I tab-switch among them pausing 1-2 seconds on each tab. > Is there any possibility that the red lines don't align with blue lines properly? Well I never make mistakes, so no, it's impossible :) But joking aside, it really seems difficult that it's wrong. I saw this pattern many times. The program that collects that data is here https://chromium-review.googlesource.com/c/chromiumos/platform2/+/775478 Tme "memd" program listens for low-memory notifications as well as dbus signals from Chrome (as discussed in #10). Those events (plus timer events) are added to a circular buffer and each is time-stamped with the current time. (There is one exception for kernel trace events, which provide their own time, but it does not apply here.)
,
Jan 19 2018
I read the code before and checked again. When /dev/chromeos-low-mem is present, the calculation logic in this memory_pressure_monitor_chromeos.cc is never used to fire critical memory pressure:
if (low_mem_file_.is_valid() &&
current_memory_pressure_level_ ==
MemoryPressureListener::MEMORY_PRESSURE_LEVEL_CRITICAL) {
current_memory_pressure_level_ =
MemoryPressureListener::MEMORY_PRESSURE_LEVEL_MODERATE;
Also in #14 I mean that it looks the only place where PurgeMemoryAndDiscardTab() (and thus your tab-discard DBus signal) is called is when memory pressure is critical. So it's really hard to imagine why a tab-discard event in the graph doesn't come with a low-mem signal. A possibility is that MemoryPressureListener is overridden by the new component "MemoryCoordinator", where critical memory pressures doesn't look to respect /dev/chromeos-low-mem . However afaik MemoryCoordinator is not enabled by default.
I've added a log at PurgeMemoryAndDiscardTab() like
// static
void TabManager::PurgeMemoryAndDiscardTab(DiscardReason reason) {
TabManager* manager = g_browser_process->GetTabManager();
manager->PurgeBrowserMemory();
manager->DiscardTab(reason);
LOG(ERROR) << "PurgeMemoryAndDiscardTab() " << TargetMemoryToFreeKB();
}
where TargetMemoryToFreeKB() prints current available memory (from /sys/kernel/mm/chromeos-low_mem/available)
The setting should be able to catch the situation where your "tab-discard" signal is raised, but available is fairly high.
However I played on my cave for a while, opening dozens of tabs and switching around. The value of (available - low_mem_margin) never exceeds 150M (usually <100M). I believe in my case low mem signal is indeed raised but some memory is reclaimed when the chrome code is executed. In your graph it's obviously not the case. What device are you using ?
I tried to patch https://chromium-review.googlesource.com/c/chromium/src/+/754301
but there's some missing file
../../chromeos/dbus/event_logger_client.cc:12:10: fatal error: 'third_party/cros_system_api/dbus/event_logger/dbus-constants.h' file not found
#include "third_party/cros_system_api/dbus/event_logger/dbus-constants.h"
so I can't try it by myself yet.
,
Jan 19 2018
Sorry---that file is not in the chromium source. I am including here for reference.
semenzato@luigi:~/chromium/src$ cat third_party/cros_system_api/dbus/event_logger/dbus-constants.h
// Copyright 2017 The Chromium OS Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef SYSTEM_API_DBUS_EVENT_LOGGER_DBUS_CONSTANTS_H_
#define SYSTEM_API_DBUS_EVENT_LOGGER_DBUS_CONSTANTS_H_
namespace event_logger {
// powerd
const char kEventLoggerInterface[] = "org.chromium.EventLogger";
const char kEventLoggerServicePath[] = "/org/chromium/EventLogger";
const char kEventLoggerServiceName[] = "org.chromium.EventLogger";
// Signal emitted for chrome events of interest for the purpose of logging
// related information (such as kernel memory manager activity at the time of a
// tab discard).
const char kChromeEvent[] = "ChromeEvent";
} // namespace event_logger
#endif // SYSTEM_API_DBUS_EVENT_LOGGER_DBUS_CONSTANTS_H_
,
Feb 16 2018
,
May 7 2018
,
May 11 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/54df2464f3fa5aa7344553e715dcd435c6b69c96 commit 54df2464f3fa5aa7344553e715dcd435c6b69c96 Author: Luigi Semenzato <semenzato@chromium.org> Date: Fri May 11 19:10:56 2018 metrics: memd: a fine-grain memory stat collection daemon Memd collects fine-grained memory stats around events of interest (such as browser tab discards) and saves them in a queue of "clip files", to be uploaded with other logs. It has two modes of operation. When memory pressure is low, it wakes up every 2 seconds and checks that pressure is still low. If not, it goes into fast-poll mode, where it collects stats in a circular buffer 10 times per second. When an "interesting" event happens, it dumps those stats into a file, limited to 5 seconds before and 5 seconds after the event. The output files (called "clip files") are also rotated, so only the most recent 20 are kept. They are stored in /var/log/memd and are uploaded (in a later CL) together with other logs in feedback reports. The daemon minimizes the number of syscalls by keeping system files open and using pread() instead of seek() + read(). Here are some performance stats in fast-poll mode: device caroline cyan cpu time/elapsed time (ratio) 0.00207846 0.00558685 user/system time (ticks) 71/335 546/4248 Since syscalls are already optimized, and user time is a small fraction of system time, further optimization is unlikely to help. If this is too high, the sampling rate could be reduced to 5 samples/second. The power_Idle autotest was used to measure times spent in various C states with and without the daemon running in fast-poll mode. This is the result on a cyan. Without memd: cpuidle_C0 5.733226548 cpuidle_C1-CHT 2.161165429 cpuidle_C6N-CHT 1.312466693 cpuidle_C6S-CHT 11.53713899 cpuidle_C7-CHT 22.70593436 cpuidle_C7S-CHT 56.55006798 cpupkg_C0_C1 17.49514512 cpupkg_C6 82.50485488 With memd: cpuidle_C0 5.826605515 cpuidle_C1-CHT 1.304592293 cpuidle_C6N-CHT 1.146455368 cpuidle_C6S-CHT 11.89613166 cpuidle_C7-CHT 28.10783807 cpuidle_C7S-CHT 51.7183771 cpupkg_C0_C1 17.12395914 cpupkg_C6 82.87604086 These numbers suggest that the impact on power will be minimal. For a functional test, run integ/memd-tester.sh from the top directory (metrics/memd). BUG=chromium:729335 TEST=integ/memd-tester on workstation Change-Id: I12a01e73efed2d845cce9567c0a5d832bd4534c5 Reviewed-on: https://chromium-review.googlesource.com/990434 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Sonny Rao <sonnyrao@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [add] https://crrev.com/54df2464f3fa5aa7344553e715dcd435c6b69c96/metrics/memd/src/main.rs [add] https://crrev.com/54df2464f3fa5aa7344553e715dcd435c6b69c96/metrics/memd/Cargo.lock [add] https://crrev.com/54df2464f3fa5aa7344553e715dcd435c6b69c96/metrics/memd/integ/memd-tester.sh [add] https://crrev.com/54df2464f3fa5aa7344553e715dcd435c6b69c96/metrics/memd/Cargo.toml
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/76520b498bce019c8243c90eb03b9a9651859feb commit 76520b498bce019c8243c90eb03b9a9651859feb Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 01:45:32 2018 metrics: memd: remove regex dependency env_logger by default depends on regex, which depends on a lot of other packages. This reduces the size of the executable from 1.4 to 0.6 MB. This also accidentally bumped the version number for a number of other dependencies. Maybe I should freeze them all in Cargo.toml. BUG=chromium:729335 TEST=tested upstart with start/stop/reboot Change-Id: I3d94c35c50e4cae1b2853b06665fc6415b2a9166 Reviewed-on: https://chromium-review.googlesource.com/1063041 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/76520b498bce019c8243c90eb03b9a9651859feb/metrics/memd/Cargo.lock [modify] https://crrev.com/76520b498bce019c8243c90eb03b9a9651859feb/metrics/memd/Cargo.toml
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/e56894fea6e462d503749b8c8df15dc204cd81c6 commit e56894fea6e462d503749b8c8df15dc204cd81c6 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 01:45:32 2018 metrics: memd: treat low-memory notifications as "interesting" This ensures that a clip is collected whenever a low-mem notification fires even if no tab is subsequently discarded (if, for instance, by the time the tab discarder is running, enough memory has been freed, by swapping or other means). BUG=chromium:729335 TEST=memd-tester still passes Change-Id: If00d8cd491d4567b70a60519fcdf6446716b8ddc Reviewed-on: https://chromium-review.googlesource.com/1063169 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Sonny Rao <sonnyrao@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/e56894fea6e462d503749b8c8df15dc204cd81c6/metrics/memd/src/main.rs
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/9cdc79abeb2b75bb8ecc24c503beb8b937e3a379 commit 9cdc79abeb2b75bb8ecc24c503beb8b937e3a379 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 01:45:33 2018 metrics: memd: correctly convert timeval fields Don't make assumptions on the size of the timeval fields (which is different on different platforms) and instead convert directly to their types. BUG=chromium:729335 TEST=compiles for kevin and caroline Change-Id: I51877ef1a0c94c817bfc3dbbb77e783172297a2b Reviewed-on: https://chromium-review.googlesource.com/1064314 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/9cdc79abeb2b75bb8ecc24c503beb8b937e3a379/metrics/memd/src/main.rs
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/f1320cb1f286178d0d7642e85b0e0866b8304fe2 commit f1320cb1f286178d0d7642e85b0e0866b8304fe2 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 01:45:33 2018 metrics: memd: use syslog crate Logging via syslog adds about 100k to the binary size (330k to 440k on x86 and 290k to 380k on arm). BUG=chromium:729335 TEST=none Change-Id: I94268a20eb89471a1b6dd49d33a644372fef28d2 Reviewed-on: https://chromium-review.googlesource.com/1064691 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [modify] https://crrev.com/f1320cb1f286178d0d7642e85b0e0866b8304fe2/metrics/memd/Cargo.lock [modify] https://crrev.com/f1320cb1f286178d0d7642e85b0e0866b8304fe2/metrics/memd/src/main.rs [modify] https://crrev.com/f1320cb1f286178d0d7642e85b0e0866b8304fe2/metrics/memd/Cargo.toml
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/2672924102bd51bf7f5897651e5854eec91a24ea commit 2672924102bd51bf7f5897651e5854eec91a24ea Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 01:45:34 2018 metrics: memd: fix parsing bug Parsing of /proc/loadavg was broken because the input could contain extra spaces. This makes the parser more lenient. Also provides more info on errors. BUG=chromium:729335 TEST=ran integration test and ran on device Change-Id: I51a5b25dc9c84cce8494bccd9ca10f4798d6bc00 Reviewed-on: https://chromium-review.googlesource.com/1066453 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [modify] https://crrev.com/2672924102bd51bf7f5897651e5854eec91a24ea/metrics/memd/src/main.rs
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/98ebc58fa5cfd33cc439233bbf45e0f45b6f5bab commit 98ebc58fa5cfd33cc439233bbf45e0f45b6f5bab Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 04:56:59 2018 metrics: memd: fix unwraps I noticed I still had a couple of incorrect unwrap() calls. I changed one to ?, the others to expect(). There are still another couple of unwraps left, but those should be OK because they are expected to succeed---it's an internal bug if they fail. BUG=chromium:729335 TEST=none Change-Id: I93ceb611fb82ff862bec08b59d69dcffe863a222 Reviewed-on: https://chromium-review.googlesource.com/1068129 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Stephen Barber <smbarber@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/98ebc58fa5cfd33cc439233bbf45e0f45b6f5bab/metrics/memd/src/main.rs
,
May 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/c6f9303a421bf01e286a363a4ca68d19880dc7e0 commit c6f9303a421bf01e286a363a4ca68d19880dc7e0 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed May 23 04:56:59 2018 metrics: memd: refactor file open and add filename to errors For performance, std::fs::File does not include path names in errors. But this program DIES on any error, so we don't need no stinking performance. BUG=chromium:729335 TEST=compiled, run on target device with minijail-induced errors Change-Id: I71b9c7fa318b034c147de2ec06dff4780af210be Reviewed-on: https://chromium-review.googlesource.com/1068130 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/c6f9303a421bf01e286a363a4ca68d19880dc7e0/metrics/memd/src/main.rs
,
Jun 7 2018
Hi Luigi, We just found that a problem in TabManager could be the root cause of what you observed in at #3 (crbug/850545).
,
Jun 8 2018
Oh yes, I saw that! It's very exciting. Well done!
,
Jun 19 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/d46f95c2baaaeb0305478f6bd46476e50c216c95 commit d46f95c2baaaeb0305478f6bd46476e50c216c95 Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Jun 19 21:38:38 2018 metrics: memd: boot script This adds /etc/init/memd.conf and two seccomp files. The seccomp policies were generated by running the daemon on all code paths (except the error paths) while recording syscalls with strace, then removing spurious entries. BUG=chromium:729335 TEST=start/stop memd on caroline and kevin Change-Id: I58a7a3ffe35049c24fbc42b502db0514fa967641 Reviewed-on: https://chromium-review.googlesource.com/1069729 Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [add] https://crrev.com/d46f95c2baaaeb0305478f6bd46476e50c216c95/metrics/memd/init/memd-seccomp-amd64.policy [add] https://crrev.com/d46f95c2baaaeb0305478f6bd46476e50c216c95/metrics/memd/init/memd.conf [add] https://crrev.com/d46f95c2baaaeb0305478f6bd46476e50c216c95/metrics/memd/init/memd-seccomp-arm.policy
,
Jun 19 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/31de6f01db4851fd27823158035466b40d3441c4 commit 31de6f01db4851fd27823158035466b40d3441c4 Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Jun 19 21:38:39 2018 metrics: memd: integrate and robustify testing These tests exercise the entire event loop by mocking time, which guarantees reproducibility and makes the test run faster (i.e. the mocked sleep() and select() with timeout don't actually do any waiting). Time is mocked by a "timer" trait which implements select(), sleep(), and now(). The "mock timer" implementation manages a list of "external" events that must be delivered at specified times. Infinite CPU speed is assumed, so the events are delivered only during sleep() or select() with a timeout. Piggy-back a small bug fix: clip files need to be truncated when clobbered. Also add ftruncate() to the seccomp whitelist and clean up a bit. BUG=chromium:729335 TEST=run "memd test", pass the tests. Also start on x86 and arm. Change-Id: I4b10c0a63e4c1c62446afbbef61a405c0c3153fb Reviewed-on: https://chromium-review.googlesource.com/1097950 Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [delete] https://crrev.com/d46f95c2baaaeb0305478f6bd46476e50c216c95/metrics/memd/integ/memd-tester.sh [modify] https://crrev.com/31de6f01db4851fd27823158035466b40d3441c4/metrics/memd/init/memd-seccomp-amd64.policy [modify] https://crrev.com/31de6f01db4851fd27823158035466b40d3441c4/metrics/memd/src/main.rs [modify] https://crrev.com/31de6f01db4851fd27823158035466b40d3441c4/metrics/memd/init/memd-seccomp-arm.policy
,
Jun 23 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/2c1f0bbff7768cb4d13cc0af9288ae9fc39e3dc2 commit 2c1f0bbff7768cb4d13cc0af9288ae9fc39e3dc2 Author: Luigi Semenzato <semenzato@chromium.org> Date: Sat Jun 23 03:19:31 2018 chromeos-base/memd: add memd ebuild This adds an ebuild for memd. I have also uploaded the needed crates to chromeos-localmirror. BUG=chromium:729335 TEST={FEATURES=test,}emerge-cyan memd; emerge-kevin memd; ran on both devices CQ-DEPEND=CL:1097950 Change-Id: I26af9f5bbf81ca1fa0141bc20f66e7a0eaea7ca2 Reviewed-on: https://chromium-review.googlesource.com/1063256 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [add] https://crrev.com/2c1f0bbff7768cb4d13cc0af9288ae9fc39e3dc2/chromeos-base/memd/memd-9999.ebuild
,
Jun 26 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/379e01218b601b03d8febf7870855f23c1140128 commit 379e01218b601b03d8febf7870855f23c1140128 Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Jun 26 00:20:23 2018 metrics: memd: split test code into separate module Moving the test code into a separate file helps achieve two goals: the main program file becomes smaller and more readable, and the test code is compiled only when running the unit tests. The separation is not perfect: the main program code still has some knowledge of testing (for instance when building the Paths object), but it's better than before. BUG=chromium:729335 TEST=ran "cargo build" and "cargo test" Change-Id: I2f3515243a0a7959005677d11c34bffdc0b4d3ce Reviewed-on: https://chromium-review.googlesource.com/1112669 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [add] https://crrev.com/379e01218b601b03d8febf7870855f23c1140128/metrics/memd/src/vmstat_content [add] https://crrev.com/379e01218b601b03d8febf7870855f23c1140128/metrics/memd/src/zoneinfo_content [add] https://crrev.com/379e01218b601b03d8febf7870855f23c1140128/metrics/memd/src/test.rs [modify] https://crrev.com/379e01218b601b03d8febf7870855f23c1140128/metrics/memd/src/main.rs
,
Jun 26 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/5687887ea07d39aae1b51f99563684c4b23c930e commit 5687887ea07d39aae1b51f99563684c4b23c930e Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Jun 26 04:00:55 2018 metrics: memd: add missing Manifest BUG=chromium:729335 TEST=none Change-Id: I8806e70d81b51df2beda2dbb7afc0e5f0edbcb8a Reviewed-on: https://chromium-review.googlesource.com/1114190 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [add] https://crrev.com/5687887ea07d39aae1b51f99563684c4b23c930e/chromeos-base/memd/Manifest
,
Jun 28 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/third_party/autotest/+/45a0694044bfed0ec0964cb746cff1ae503777c5 commit 45a0694044bfed0ec0964cb746cff1ae503777c5 Author: Luigi Semenzato <semenzato@chromium.org> Date: Thu Jun 28 05:06:59 2018 platform_ToolchainOptions: whitelist /usr/bin/memd BUG=chromium:729335 TEST=none Change-Id: Ia8878a50463b07a2a87ef2c4e7523befef25baf0 Reviewed-on: https://chromium-review.googlesource.com/1116353 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/45a0694044bfed0ec0964cb746cff1ae503777c5/client/site_tests/platform_ToolchainOptions/libgcc_whitelist
,
Jun 28 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/b0f79d073c22145d953765cae3be91897021adf2 commit b0f79d073c22145d953765cae3be91897021adf2 Author: Luigi Semenzato <semenzato@chromium.org> Date: Thu Jun 28 05:07:00 2018 memd: deploy memory daemon Deploy everywhere except "embedded" and "minimal" profiles. BUG=chromium:729335 TEST=built for an amd64 target CQ-DEPEND=CL:1114190, CL:1116353 Change-Id: I105720804f8e9ff361006cb72bcb96a462ca1fd6 Reviewed-on: https://chromium-review.googlesource.com/1112338 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [modify] https://crrev.com/b0f79d073c22145d953765cae3be91897021adf2/profiles/features/minimal/make.defaults [modify] https://crrev.com/b0f79d073c22145d953765cae3be91897021adf2/profiles/targets/chromeos/make.defaults [modify] https://crrev.com/b0f79d073c22145d953765cae3be91897021adf2/profiles/features/embedded/make.defaults [rename] https://crrev.com/b0f79d073c22145d953765cae3be91897021adf2/virtual/target-chromium-os/target-chromium-os-1-r100.ebuild [modify] https://crrev.com/b0f79d073c22145d953765cae3be91897021adf2/virtual/target-chromium-os/target-chromium-os-1.ebuild
,
Jun 28 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/39cc371b22066adbf0e6c5f4054c84941b4a5483 commit 39cc371b22066adbf0e6c5f4054c84941b4a5483 Author: Luigi Semenzato <semenzato@chromium.org> Date: Thu Jun 28 05:06:58 2018 memd: do not run unit tests in non-x86 builds BUG=chromium:729335 TEST=FEATURES=test emerge-kevin memd Change-Id: Ibe531d1cc38d9294d958ae4d63dcf7659c985037 Reviewed-on: https://chromium-review.googlesource.com/1115339 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/39cc371b22066adbf0e6c5f4054c84941b4a5483/chromeos-base/memd/memd-9999.ebuild
,
Jun 28 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/7f00dfb8d2eb69e76c95199ced094a776a60b26f commit 7f00dfb8d2eb69e76c95199ced094a776a60b26f Author: Luigi Semenzato <semenzato@chromium.org> Date: Thu Jun 28 05:07:12 2018 debugd: add collection of memd logs BUG=chromium:729335 TEST=none Change-Id: Ie78613c1f5404efce61e5bf38f3559ec5f3fcc21 Reviewed-on: https://chromium-review.googlesource.com/1115703 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Ben Chan <benchan@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/7f00dfb8d2eb69e76c95199ced094a776a60b26f/debugd/src/log_tool.cc
,
Jul 10
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/third_party/autotest/+/b70dd180671bf0651921b926427b8372b4126478 commit b70dd180671bf0651921b926427b8372b4126478 Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Jul 10 00:59:30 2018 security_SandboxedServices: add memd to baseline BUG=chromium:729335 TEST=./bin/autotest_client tests/security_SandboxedServices/control Change-Id: I31fb6e5ef5255e26ed50efb3c5a5105d336d768f Reviewed-on: https://chromium-review.googlesource.com/1128500 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/b70dd180671bf0651921b926427b8372b4126478/client/site_tests/security_SandboxedServices/baseline
,
Jul 15
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/f4efe08ab35905cde61ec77c1f47b85e7864b40d commit f4efe08ab35905cde61ec77c1f47b85e7864b40d Author: Mike Frysinger <vapier@chromium.org> Date: Sun Jul 15 01:11:43 2018 memd: disable for R69 It's pretty late in the branch cycle, so pull memd out as it hasn't had enough testing just yet. We can re-enable for R70. BUG= chromium:860034 ,chromium:729335 TEST=precq passes Change-Id: Ib8f97c57689ce5d14bf24120643df419412dc7e0 Reviewed-on: https://chromium-review.googlesource.com/1136874 Commit-Ready: Mike Frysinger <vapier@chromium.org> Tested-by: Mike Frysinger <vapier@chromium.org> Reviewed-by: Sonny Rao <sonnyrao@chromium.org> [modify] https://crrev.com/f4efe08ab35905cde61ec77c1f47b85e7864b40d/profiles/targets/chromeos/make.defaults
,
Aug 3
This bug has an owner, thus, it's been triaged. Changing status to "assigned".
,
Aug 24
I should probably own this at least for now.
,
Aug 30
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform/system_api/+/a71cda985de20c747a6525de0fac2dc95766bd6d commit a71cda985de20c747a6525de0fac2dc95766bd6d Author: Luigi Semenzato <semenzato@chromium.org> Date: Thu Aug 30 04:05:46 2018 system_api: dbus: add metrics_event protobuffer This is used for signaling memory pressure events from Chrome to memd (or other interested party). BUG=chromium:729335 TEST=compiles and runs correctly with chromium changes that use it Change-Id: I00aee977045977d606e3f09eb8a4215c54ea5ee2 Reviewed-on: https://chromium-review.googlesource.com/1186024 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Dan Erat <derat@chromium.org> [add] https://crrev.com/a71cda985de20c747a6525de0fac2dc95766bd6d/dbus/metrics_event/service.proto
,
Aug 31
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/e15256b1402e6bf98c3490b96745fbc1aa5b748c commit e15256b1402e6bf98c3490b96745fbc1aa5b748c Author: Luigi Semenzato <semenzato@chromium.org> Date: Fri Aug 31 12:23:47 2018 chromeos-base/memd: adjust ebuild for use of protobufs. These changes are needed after adding the use of protobufs to memd, for parsing metrics event signals from Chrome. BUG=chromium:729335 TEST=compiles Change-Id: Ie608b2a052efa61d67e7200816df44669c452990 Reviewed-on: https://chromium-review.googlesource.com/1194651 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Zach Reizner <zachr@chromium.org> [modify] https://crrev.com/e15256b1402e6bf98c3490b96745fbc1aa5b748c/chromeos-base/memd/memd-9999.ebuild [modify] https://crrev.com/e15256b1402e6bf98c3490b96745fbc1aa5b748c/chromeos-base/memd/Manifest
,
Sep 1
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform/system_api/+/ac2aa599abbd9bb353914648e83de934a15b3682 commit ac2aa599abbd9bb353914648e83de934a15b3682 Author: Luigi Semenzato <semenzato@chromium.org> Date: Sat Sep 01 21:43:34 2018 system_api: dbus: rename metrics_event proto file Naming the protobuf specification "service.proto" causes conflicts in the chrome build. This may be related to https://github.com/protocolbuffers/protobuf/issues/2520 namespace protobuf_service_2eproto { void InitDefaults() { Naming the file "service.proto" makes the namespace contain the word "service", which happens to be used by another protobuf for vm_concierge. So I get a link-time multiple definition. BUG=chromium:729335 TEST=compiles and runs correctly with chromium changes that use it Change-Id: Ic83489d3bab15f627288e44c323e5032da76d874 Reviewed-on: https://chromium-review.googlesource.com/1199965 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Dan Erat <derat@chromium.org> [rename] https://crrev.com/ac2aa599abbd9bb353914648e83de934a15b3682/dbus/metrics_event/metrics_event.proto
,
Sep 5
(Pardon me. Doing a bulk audit for CLs that missed R70 for nocturne: https://crosland.corp.google.com/log/11021.0.0..11035.0.0 This CL was found in the "might be helpful for nocturne" list). CL in #43 went in before R70 branch point, but #44 and #45 came later. Is this intended for a Merge-Request to M70?
,
Sep 5
,
Sep 5
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform/system_api/+/41fa3011067832b0b4341de834cd7cd341099738 commit 41fa3011067832b0b4341de834cd7cd341099738 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed Sep 05 12:11:43 2018 system_api: service_constants.h: add MetricsEvent This adds constants for the MetricsEventService. Ideally this would have gone in with the addition of metrics_event.proto, but I had my chrome and cros sources out of sync. BUG=chromium:729335 TEST=chromium compiles (cros clones these in Rust code) Change-Id: I2407619196dd6fe75e3143caa932bb443ad5fffc Reviewed-on: https://chromium-review.googlesource.com/1204910 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Dan Erat <derat@chromium.org> [modify] https://crrev.com/41fa3011067832b0b4341de834cd7cd341099738/dbus/service_constants.h
,
Sep 5
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/1f38337baf092db94acc789741efaa243a1fb836 commit 1f38337baf092db94acc789741efaa243a1fb836 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed Sep 05 23:11:12 2018 chromeos-base/memd: add dependency on chromeos-base/system_api BUG=chromium:729335 TEST=none Change-Id: I3231a8e08db5e4f2ccbdd2de57033bcb8b30463a Reviewed-on: https://chromium-review.googlesource.com/1199445 Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/1f38337baf092db94acc789741efaa243a1fb836/chromeos-base/memd/memd-9999.ebuild
,
Sep 7
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/a179f27585302bacbcca81263479ba3e9c28ce36 commit a179f27585302bacbcca81263479ba3e9c28ce36 Author: Luigi Semenzato <semenzato@chromium.org> Date: Fri Sep 07 21:06:41 2018 Add metrics event reporting for Chrome OS This adds a mechanism to report statistically-interesting events to Chrome OS listeners. A Chrome OS daemon (such as the memory daemon "memd") can register as a client of the Metrics Event Service, and receive signals on the occurrence of related events (for instance, a tab discard). This CL adds the supporting machinery and its first use, signaling of a tab discard. CQ-DEPEND=CL:1199965 BUG=chromium:729335 TEST=ran on DUT, verified that D-Bus signal is sent as expected Change-Id: I6eddf4b0d18aa4c8a3fd4aab7f0770d87ddc7aeb Reviewed-on: https://chromium-review.googlesource.com/1062078 Commit-Queue: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Steven Bennetts <stevenjb@chromium.org> Reviewed-by: Dan Erat <derat@chromium.org> Cr-Commit-Position: refs/heads/master@{#589639} [modify] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/DEPS [modify] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chrome/browser/chromeos/BUILD.gn [modify] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chrome/browser/chromeos/chrome_browser_main_chromeos.cc [add] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chrome/browser/chromeos/dbus/metrics_event_service_provider.cc [add] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chrome/browser/chromeos/dbus/metrics_event_service_provider.h [add] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chrome/browser/chromeos/dbus/org.chromium.MetricsEventService.conf [modify] https://crrev.com/a179f27585302bacbcca81263479ba3e9c28ce36/chromeos/BUILD.gn
,
Sep 8
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/22f4b09a56a133a476e13eb71a58cd3de356b844 commit 22f4b09a56a133a476e13eb71a58cd3de356b844 Author: Luigi Semenzato <semenzato@chromium.org> Date: Sat Sep 08 05:27:45 2018 metrics: memd: plumbing D-Bus event notifications from Chrome Register for notifications from the Metrics Event Service provided by Chrome. This service offers no methods, only sends signals for events that have a large impact on memory usage, such as tab discards and tab switches. The changes include: - adding build.rs to generate Rust for protobuf; - parsing D-Bus payload using protobuf code; - changing process_chrome_events to return a list of events (each with its own timestamp) rather than only a count; - fixing the unit tests to include time stamps in the mock chrome events. - making the output routine ignore possibly incorrect time stamps from external sources CQ-DEPEND=CL:1194651,CL:1199445,CL:1199965 BUG=chromium:729335 TEST=unit test, ran on device, verified tab discards work as expected Change-Id: Ica9805ec134d40d44984de3d5276b9fae9257a86 Reviewed-on: https://chromium-review.googlesource.com/1188796 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [add] https://crrev.com/22f4b09a56a133a476e13eb71a58cd3de356b844/metrics/memd/build.rs [modify] https://crrev.com/22f4b09a56a133a476e13eb71a58cd3de356b844/metrics/memd/Cargo.lock [modify] https://crrev.com/22f4b09a56a133a476e13eb71a58cd3de356b844/metrics/memd/src/main.rs [modify] https://crrev.com/22f4b09a56a133a476e13eb71a58cd3de356b844/metrics/memd/src/test.rs [modify] https://crrev.com/22f4b09a56a133a476e13eb71a58cd3de356b844/metrics/memd/Cargo.toml
,
Sep 24
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/3d29b40bd87d3adf42730d2dbd4822f21e0cdaea commit 3d29b40bd87d3adf42730d2dbd4822f21e0cdaea Author: Luigi Semenzato <semenzato@chromium.org> Date: Mon Sep 24 17:20:55 2018 histograms.xml: add Network.Shill.CumulativeTimeOnLine.* Implementation CL: https://chromium-review.googlesource.com/c/aosp/platform/system/connectivity/shill/+/1220430 BUG=chromium:729335 TEST=none Change-Id: I00fe89328d9ec0f910b354733fb1914838c5d75c Reviewed-on: https://chromium-review.googlesource.com/1234696 Commit-Queue: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Ilya Sherman <isherman@chromium.org> Reviewed-by: Robert Kaplow (sloooow) <rkaplow@chromium.org> Cr-Commit-Position: refs/heads/master@{#593579} [modify] https://crrev.com/3d29b40bd87d3adf42730d2dbd4822f21e0cdaea/tools/metrics/histograms/histograms.xml
,
Sep 26
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform/system_api/+/98bc2a6d4b6c83b6dcae8fe50101aef1b75d1a13 commit 98bc2a6d4b6c83b6dcae8fe50101aef1b75d1a13 Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed Sep 26 17:32:14 2018 system_api: add anomaly event service interface This also adds new event types to the metrics event protobuf, and adds generation of C++ sources for use by the anomaly collector. BUG=chromium:729335 TEST=compiles, links, and runs with anomaly_collector Change-Id: I330d9103752c9db58c6db824f69e25d5cc6bd180 Reviewed-on: https://chromium-review.googlesource.com/1239736 Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [modify] https://crrev.com/98bc2a6d4b6c83b6dcae8fe50101aef1b75d1a13/dbus/metrics_event/metrics_event.proto [modify] https://crrev.com/98bc2a6d4b6c83b6dcae8fe50101aef1b75d1a13/dbus/service_constants.h [modify] https://crrev.com/98bc2a6d4b6c83b6dcae8fe50101aef1b75d1a13/system_api.gyp
,
Sep 26
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/overlays/chromiumos-overlay/+/21a9b57ba39bb8010f891284680081957dfe415d commit 21a9b57ba39bb8010f891284680081957dfe415d Author: Luigi Semenzato <semenzato@chromium.org> Date: Wed Sep 26 17:32:14 2018 chromeos-base/system_api: add metrics_event protobuf. The metrics_event protobuf is used by the anomaly detector to send a D-Bus signal when it detects that the kernel has attempted an Out-of-memory kill (OOM kill). CQ-DEPEND=CL:1239736 BUG=chromium:729335 TEST=compiled Change-Id: I9722752c1a28076535cf69d465bc03a990a564a5 Reviewed-on: https://chromium-review.googlesource.com/1236995 Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Luigi Semenzato <semenzato@chromium.org> [modify] https://crrev.com/21a9b57ba39bb8010f891284680081957dfe415d/chromeos-base/system_api/system_api-9999.ebuild
,
Oct 2
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/a457cc7a65c8cd23572df023e7eaf88e3f219a4e commit a457cc7a65c8cd23572df023e7eaf88e3f219a4e Author: Luigi Semenzato <semenzato@chromium.org> Date: Tue Oct 02 16:17:40 2018 crash: add detection and reporting of OOM kills Enhance the anomaly collector so that it detects kernel OOM kills, ahd correspondingly sends D-Bus signals. This is mainly for memd, which collects and correlates such events. BUG=chromium:729335 TEST=ran on DUT, verified content of emitted signals Change-Id: I1f1055ff255892f646edc81d39c3bce14b2acc0e Reviewed-on: https://chromium-review.googlesource.com/1224637 Commit-Ready: Luigi Semenzato <semenzato@chromium.org> Tested-by: Luigi Semenzato <semenzato@chromium.org> Reviewed-by: Dan Erat <derat@chromium.org> [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/anomaly_collector_test.sh [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/README.md [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/anomaly_collector.l [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/anomaly_collector.h [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/anomaly_collector.cc [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/BUILD.gn [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/metrics/memd/src/main.rs [modify] https://crrev.com/a457cc7a65c8cd23572df023e7eaf88e3f219a4e/crash-reporter/docs/collectors.md
,
Oct 17
,
Oct 17
Issue 896142 has been merged into this issue.
,
Oct 18
(Continue the discussion of bug 896142 since it's merged into this bug) Conclusion: we should not invoke memory pressure handler if the previous invocation hasn't finished. If, for any reason, the previous dispatch of TabManager::OnMemoryPressure() hasn't finished its job (e.g. will take 5 sec to finish), the memory pressure monitor will almost certainly dispatch another 4. When the first finishes its job, the other 4 dispatches will run in a bursty manner *right after* the first one and will either: 1. still find some app/tab to kill. 1.1 This could be either another tab/app is still quickly allocating memory. 1.2 The app/tab killed in the first dispatch hasn't taken effect in system memory statistics. If this is the case, we will see several dispatches has the same number of "Target memory to free". Because the later dispatches doesn't know previous kill in action, we will overkill. This will be bad for user because more than necessary tabs/apps will be discarded in a sudden. 2. cannot find some app/tab to kill. We will see from the log of listing of candidates, but no app/tab is killed because target memory to free is already negative. It just wastes a lot of cycles on the UI thread. This also has a side effect of the first-kill latency metric: if 1.1 happens, we will send out first-kill latencies ike 5, 4, 3, 2, 1 seconds, but these latencies actually overlap in time, which doesn't make much sense.
,
Oct 23
|
||||||||||||||
►
Sign in to add a comment |
||||||||||||||
Comment 1 by sonnyrao@chromium.org
, Jun 23 2017