...in ActivityManagerService.updateLruProcessInternalLocked on bluetooth
Add more debug output to help track down what is going on.
Also fix a little problem where, when a service ANRs, if you ask to
wait and it still wasn't responding, the ANR dialog wouldn't be
shown again.
Change-Id: I5be2b1705a0a39ca2992624ae683945c5f38065d
...be uncached and too large
When the device is in a low RAM state, when we go to pull a cached
process out to use for some background operation, we can now kill
the current process if we consider its size to be too large.
Note that the current implementation for killing processes is to
just use the same killUnneededProcessLocked() method that we already
have for other things like too many cached processes. This is a
little wrong here, though, because in this case we are at the
point where the caller is actually looking for a process to use.
This current code is not actually removing or cleaning up the
process, so we still need to return the now killed ProcessRecord
and let things fall out from there, which typically means the caller
trying to make an IPC on it and failing and falling into its "oh
no the process died unexpectedly" path. All code using this
*should* be able to handle this correctly, anyway, since processes
really can be killed at any time.
At some point we may to make this implementation cleaner, where it
actually tears down the process right in the call and returns a
null ProcessRecord. That is very dangerous however (we'd need to
go through all paths into this to make sure they are going to be
okay with process state changing on them like that), and I'm not
sure it is really worthwhile. This intention is that killing
processes like this is unusual, due to processes being too large,
and anyway as I wrote all of our incoming code paths must already
be able to handle the process being killed at this point and one
could argue this is just another way to excercise those code paths.
Really, the main negative to this is that we will often have spam
in the log with exceptions about processes dying unexpectedly.
If that is the only issue, we could just add some conditions to
quiet that up at in this case.
We don't want to compute the size of the process each time we try
to evaluate it here (it takes 10s or ms to do so), so there is now
a new field associated with the process to give us the last pss
size we computed for it while it was in the cached state.
To be able to have better cached pss data when we now need it, the
timing for computing process pss has been tuned to use a much
shorter delay for the situations when the process has first switch
into a new state. This may result in us having a fair amount more
pss data overall, which is good, as long as it doesn't cause us to
be computing pss excessively and burning cpu.
Procstats now also has new state to keep track of the number of
times each process has been killed by this new system, along with
the min, avg, max pss of all the times it has happened. This has
slightly changed the checkin format to include this additional data
at the end of pkgkills/prockills lines.
Other changes here:
- Fixed a problem where GPU RAM was not being seen when dumping
the full RAM details of a process. This was because in that
case the system would ask the process to compute its own MemInfo,
which it returned, but the process doesn't have permission to
access the files containing the GPU RAM data. So now the system
always computes the MemInfo and hands it to the app.
- Improved broadcast delays to not apply the delay if the next receiver
of the broadcast is going to run in the same process as the last
one. A situation I was seeing was an application that had two
receivers, one of which started a service; we are better off letting
the second receiver run while the service is running.
- Changed the alarm manager's TIME_TICK broadcast to be a foreground
broadcast. This really should have been anyway (it is supposed to
go out even minute, on the minute, very accurately, for UI elements
to update), and is even more important now that we are doing more
things to delay background broadcasts.
- Reworked how we maintain the LRU process list. It is now divided
into the two parts, the top always containing the processes holding
activities. This better matches the semantics we want (always try
to keep those around modulated by the LRU order we interleave with
other cached processes), and we now know whether a process is being
moved on the LRU list because of an activity operation so we can
only change the order of these activity processes when user operations
happen. Further, this just makes that common code path a lot simpler
and gets rid of all the old complexity that doesn't make sense any
more.
Change-Id: I04933ec3931b96db70b2b6ac109c071698e124eb
...while setting up a new user from settings.
We can now delay broadcasts when there are enough background services
currently starting (still set to 1 for svelte devices, 3 for normal
devices).
Add new intent flag to not allow receivers to abort broadcasts, which
I use to fix an issue with the initial BOOT_COMPLETED broadcast not
actually requesting pss data at the right time -- it can now be sent
as an ordered broadcast without the ability for the receivers to cancel
it.
Change-Id: I51155bbbabe23e187003f3e2abd7b754e55d3c95
...while setting up a new user from settings.
The delayed service start stuff was too aggressive -- it would
allow a process to be killed between the an onReceive() that calls
startService() and that service being started. This means that
apps that set up global state that they expect to remain set up
during that time could be lost.
This is the first part of a fix, which tightens up when we allow
services to be delayed. Now we will immediately start the service
as long as it currently as a process running that is not in the
cached state. (Previously we would delay if the process was in
the receiver state.)
This unfortunately means that our service start delay is much
less effective. To address that, there will be a follow-on change
to tie broadcast delivery into this to see if we can delay the
finish of a broadcast as long as there are background services
starting in that process.
Change-Id: I2bba2295d10699ee3479375bbe87114b2cbb0826
Dumb typo was clearing the wrong service array, causing
us to sometimes forget we were launching a service.
Change-Id: Ie1aba0e07d19e85a104a5985e3cead5f28a0556a
This significantly reworks the logging we do when
all cached processes are killed:
- We now collect the list of processes in-place so we
have a snapshot of exactly when the low memory situation
happened.
- In that snapshot we include the key process state: oom
adj, proc state, adj reasons.
- The report then asynchronously collects pss information
for those processes.
- The ultimate data printed to the log looks like a mix
between the "dumpsys meminfo" and "dumpsys activity"
output. This code no longer uses "dumpsys meminfo"
itself, so some of that data is no longer included,
in particular pss organized by allocation type.
In doing this, I realized that the existing code that is
supposed to run "procstats" is not currently working. And
at that point I realized, really, when we are collecting
this pss data we'd really like to include all those native
processes using ghod-only-knows how much RAM. And guess
what, we have a list of processes available in
ProcessCpuTracker.
So we now also collect and print information for native
processes, and we also do this for "dumpsys meminfo" which
really seems like a good thing when we are printing summaries
of all pss and such.
I also improved the code for reading /proc/meminfo to be
able to load all the interesting fields from there, and
am now printing that as well.
Change-Id: I9e7d13e9c07a8249c7a7e12e5433973b2c0fdc11
Added some code to the activity manager to keep track of
services that are launching and limit the number that can
be launched concurrently. This only comes into play under
specific circumstances: when the service launch is a background
request (so timing is not important) and its process is not
already running at a high priority.
In this case, we have a list of services that are currently
launching and when that gets too big we start delaying the
launch of future services until currently launching ones are
finished.
There are some important tuning parameters for this: how many
background services we allow to launch concurrently (currently
1 on low-ram devices, 3 on other devices), and how long we
wait for a background service to run before consider it to be
a more long-running service and go on to the next pending
launch (currently set to 15 seconds).
Also while in here, did some cleanup of the service code:
- A little refactoring to make per-user data cleaner.
- Switch to ArrayMap.
Change-Id: I09f372eb5e0f81a8de7c64f8320af41e84b90aa3
ProcessStats is now called ProcessCpuTracker.
ProcessTracker is now ProcessStatsService, and its inner State
class is broken out into a separate top-level ProcessStats class.
This ProcessStats is moved to the framework, so we will be able
to use it elsewhere.
Change-Id: I6a127bcb835b6b474b72647c0b99b82c2137e5c5
If a service is being executed due to a call from a background
process, there is no reason for its process to be brought up
to the foreground scheduling group to do its work. So, don't.
Change-Id: I0f4f9f2fdcc3481ca4ad2e4e3deb43b8edbbd897
When safely resetting stats after committing them, we were
mistakenly clearing the current memory state so we would lose
that total memory time until the memory state changes again.
Also improve the summary output to print percentages, which
make more sense for that display.
Change-Id: I0fe45fd78e97ec8b94976170dd42f4ed345a5899
These new constants are a better mapping to the kind of
information that procstats is wanting to collect about
processes. In doing this, the process states are tweaked
to have a bit more information that we care about for
procstats.
This changes the format of the data printed by procstats,
so the checkin version is bumped to 2. The structure is
the same, however the codes for process states have all
changed. The new codes are, in order of precedence:
p -- persistent system process.
t -- top activity; actually any visible activity.
f -- important foreground process (ime, wallpaper, etc).
b -- important background process
u -- performing backup operation.
w -- heavy-weight process (currently not used).
s -- background process running a service.
r -- process running a receiver.
h -- process hosting home/launcher app when not on top.
l -- process hosting the last app the user was in.
a -- cached process hosting a previous activity.
c -- cached process hosting a client activity.
e -- cached process that is empty.
In addition, we are now collecting uss along with pss
data for each process, so the pss checkin entries now
have three new values at the end of the min/avg/max uss
values of that process.
With this switch to using process state constants more
fundamentally, I realized that they could actually be
used by the core oom adj code to make it a lot cleaner.
So that change has been made, that code has changed quite
radically, and lost a lot of its secondary states and flags
that it used to use in its computation, now relying on
primarily the oom_adj and proc state values for the process.
This also cleaned up a few problems -- for example for
purposes of determing the memory level of the device, if a
long-running service dropped into the cached oom_adj level,
it would start being counted as a cached process and thus
make us think that the memory state is better than it is.
Now we do this based on the proc state, which always stays
as a service regardless of what is happening like this, giving
as a more consistent view of the memory state of the device.
Making proc state a more fundamentally part of the oom adj
computation means that the values can also be more carefully
tuned in semantic meaning so the value assigned to a process
doesn't tend to change unless the semantics of the process
has really significantly changed.
For example, a process will be assigned the service state
regardless of whether that services is executing operations
in the foreground, running normally, or has been dropped to
the lru list for pruning. The top state is used for everything
related to activities visible to the user: when actually on
top, visible but not on top, currently pausing, etc.
There is a new Context.BIND_SHOWING_UI added for when system
services bind to apps, to explicitly indicate that the app
is showing UI for the system. This gives us a better metric
to determine when it is showing UI, and thus when it needs
to do a memory trim when it is no longer in that state. Without
this, services could get in bad states of continually trimming.
Finally, more HashSet containers have been changed to ArraySet,
reducing the temporary iterators created for iterating over
them.
Change-Id: I1724113f42abe7862e8aecb6faae5a7620245e89
The activity manager now keeps a new "process state" for
each process, indicating the general execution and memory
state of the process. This closely follows the out-of-memory
adjustment and scheduling class that it currently tracks,
but roles these together (plus a little more info) into one
more semantically meaningful number.
This value is reported to each process as it changes, so they
can do things like tune the Dalvik garbage collector to match
the current process state.
I think I should also switch to this for process states. It
will give is more meaningful divisions of time for each process.
Also fix a problem in the activity stack where the previous
process was not being set correctly when moving between
activity stacks.
Change-Id: I598b1667dc46547f8fadae304e210c352cc9d41f
- Only log battery levels on level changes, not voltage or temp.
- Reduce the text in the start service log to only the suffix
of the service component, and a uid field.
Change-Id: I6d65e6700e021b3b005dccc90d64f36c1f97137f
We now persistent the current procstats to storage
to keep them across boots. Still need to do division
and pruning across days; right now they will just keep
collecting forever.
Also fix some bugs in the checkin output.
Change-Id: I4dd9317dbe2ee0642af8f2f0be1f2bd9c4055e80
This should save some memory on devices with limited amounts of RAM
by allowing it to share the graphics context with systemui.
Initially, adding this to the systemui process triggered a bug where
the shared process id didn't exist in the table yet. The change to
ActiveServices tries to work around this by explicitly adding the package in
attachApplicationLocked().
Change-Id: I9e97d5d69ddd15d7f16aca83a57bdf381811a13b
Re-arranged code to be more flexible, now track
state of services, dump ordered list of running
processes while memory was critical and low.
Also rename battery stats service from "batteryinfo" to "batterystats".
Change-Id: I0f4f0c8d443c49d255cb84d0fc917e8ec18b152e
We now keep track of when each process is running, batched
by the current memory status of the device. In addition,
the stats are organized by package first, and then processes
associated with each package inside of that. Finally, we
also keep track of the overall time spent in each memory
status.
This should start to actually get us to some information
we can reach some conclusions about. The total time spent
in each memory status gives us some indication of how much
we are running while memory is low; the new package organization
batched by memory status lets us see what packages have
what processes running when memory is low.
Change-Id: I389d62d39d115a846126cf354e4c20070d8f1180
Adds a platform API, and pm command. Fixes some issues with
dumping per-package data in package manager, makes battery
stats able to dump per-package state.
Change-Id: I76ee6d059f0ba17f7a7061886792b1b716d46d2d
The goal of this is to keep track of what app processes
are doing, to determine who is being abusive, when the system
is getting into memory constrained situations, and help the
user determine how to resolve this.
Right now it doesn't really do any of that, just keeps track
of how long every process has been running since boot.
Also update the activity manager to use "cached" as the terminology
for what it used to interchangeably call hidden and background
processes, and switch ProcessMap over to using ArrayMap.
Change-Id: I270b0006aab1f38e17b7d9b65728679173c343f2
Dumping per-package data is now much cleaning, and only really
prints information related to the package. Also clean up the
formatting and such of the new activity stack states.
Change-Id: Ie5f31d72b7e77b34fd9acd835f28ed1db1dff328
Yay to ArrayMap, letting me get rid of a bunch of temporary
iterators in core code paths like updateOomAdj. (Now I definitely
need an ArraySet to finish that up.)
Also clean up various other things that are doing unnecessary
allocations, clean up some debug output, make more of the debug
output respect package filtering.
Change-Id: Ib4979faf4de8c7912739bc0937c3fa9e7bfcde67
- Replace mMainStack with mFocusedStack and mStacks.
- Remove stack from ActivityRecord.
- Add stack to TaskRecord.
Change-Id: I22e9ba34b12c2bd90806b14aafe063d5a2fe66ae
Omits service name from destroyed events, since it can be derived by
looking back to the created event with the same ServiceRecord.
Change-Id: Ib7ab1031c0859437735e1fc985d58f47629b7ac4
...phase & callback API
I realized there were a few things wrong with what was there. The new
ACTION_USER_STARTING was not being sent for the first user at boot, and
there was an existing problem where ACTION_USER_STARTED was sent every
time there was a user switch.
Also improved some debug output of broadcasts to make it easier to see
what is going on in this stuff, and better reporting of why a service
couldn't be started.
Change-Id: Id8a536defbbad1f73d94a37d13762436b822fbe3
Add a new call to the activity manager for the input dispatcher
to report about any pid having an ANR. This has a new feature
where it can also tell the activity manager that it is above the
system alert layer, so the activity manager can pop its ANR dialog
on top of everything if it needs to. (Normally we don't want
these dialogs appearing on top of the lock screen.)
Also fixed some debugging stuff here and there that was useful
as I was working on this -- windows now very clearly include
their uid, various system dialogs now have titles so you know
what they are in the window manager, etc.
Change-Id: Ib8f5d29a5572542cc506e6d338599ab64088ce4e
...Forground Sometimes Doesn't Take
The main change here is a one-liner in ActiveServices to check the
uid when deciding whether to remove an item from mPendingServices.
This could cause the problem being seen -- if the same service for
two users is starting at the same time, the second one would blow
away the pending start of the first one. Unfortunately I have had
trouble reproducing the bug, so I don't know if this is actually
fixing it. It's a bug, anyway.
The reason so much has changed here is because I spread around
logging and printing of the user ID associated with operations and
objects to make it easier to debug these kind of multi-user things.
Also includes some tweaks to the oom manager to allow more background
processes (I have seen many times in logs where we thrash through
processes because the LRU list is too short), plus to compensate an
additional time-based metric for when to get rid of background processes,
plus some new logic to try to help things like Chrome keep around
their service processes.
Change-Id: Icda77fb2a1dd349969e3ff2c8fff0f19b40b31d3
...service com.android.systemui.SystemUIService: java.lang.NullPointerException
- Don't acquire the activity manager lock in handleIncomingUser(),
there is really no need to do so.
- Rework the settings provider client side cache code to not hold
locks while calling into the provider.
I also changed the way the settings provider uses system properties
so that there is one property for all users. We can't do one per
user, since the system property name space is limited with a fixed
size. And we don't really need to do that; the worse that happens
by combining all users is that if one running user changes one of its
settings, all other running users will think they need to reload
settings when they go to fetch them next.
Change-Id: I13b90b832310d117eb6d721aacd122cfba7d749a
Now we default to the current user instead of user 0 for most commands
(except where we can do the command for all users).
Many more commands take a user argument: force-stop, kill, profile,
dumpheap.
Improved help text.
Change-Id: I719a13b4d31b668f57ca21e51d7043ac3e0d4e1b
Bug #7132226: Can't start instrumentation due to ActivityManagerService crash
Bug #6912004: tap on gmail notification sends me to home screen
Change-Id: I824128b01f368de95dee288f8e49039b84479a7e
- New (hidden) isUserRunning() API.
- Maintain LRU list of visited users.
- New FLAG_IS_DATA_ONLY for ApplicationInfo.
- Clean up pending intent records when force-stopping a user (or package).
(Also fixes bug #6880627: PendingIntent.getService() returns stale
intent of force stopped app)
- Fix force-stopping when installing an app to do the force-stop across
all users for that app.
- When selecting which processes to kill during a force stop, do this
based on the actual packages loaded in the process, not just process
name matching.
- You can now use --user option in am when starting activities, services,
and instrumentation.
- The am --user option accepts "current" and "all" as arguments.
- The pm uninstall command now uninstalls for all users, so it matches
the semantics of the install command.
- PhoneWindowManager now explicitly says to start home in the current
user.
- Activity manager call to retrieve the MIME type from a content provider
now takes a user argument, so it will direct this to the proper user.
- The package manager uninstall paths are now implemented around
PackageSetting, not PackageParser.Package. This allows them to work
even if the application's apk has been removed (in which case it only
exists as a PackageSetting, not the PackageParser.Package parsed from
the apk).
Change-Id: I3522f6fcf32603090bd6e01cc90ce70b6c5aae40
Replaced all remaining places that used it with explicit user
specification.
While doing this, I ran into stuff that was creating PendingIntent
objects (that now need to specify the explicit user they are for),
which are also posting notifications... but have no way to specify
the user for the notification.
So the notification manager in the system process now also gets a
formal concept of a user associated with the notification, which
is passed in to all the necessary aidl calls. I also removed the
old deprecated aidl interface for posting/cancelling notifications,
since we now always need a user supplied.
There is more work that needs to be done here, though. For example
I think we need to be able to specify USER_ALL for a notification that
should be shown to all users (such as low storage or low battery).
Along with that, the PendingIntent creation needs to be tweaked to
be able to handle USER_CURRENT by evaluating the user at the point the
pending intent is sent.
That's for another change, however.
Change-Id: I468e14dce8def0e13e0870571e7c31ed32b6310c
You can now use ALL and CURRENT when sending broadcasts, to specify
where the broadcast goes.
Sticky broadcasts are now correctly separated per user, and registered
receivers are filtered based on the requested target user.
New Context APIs for more kinds of sending broadcasts as users.
Updating a bunch of system code that sends broadcasts to explicitly
specify which user the broadcast goes to.
Made a single version of the code for interpreting the requested
target user ID that all entries to activity manager (start activity,
send broadcast, start service) use.
Change-Id: Ie29f02dd5242ef8c8fa56c54593a315cd2574e1c
The activity manager now keeps track of which users are running.
Initially, only user 0 is running.
When you switch to another user, that user is started so it is
running. It is only at this point that BOOT_COMPLETED is sent
for that user and it is allowed to execute anything.
You can stop any user except user 0, which brings it back to the
same state as when you first boot the device. This is also used
to be able to more cleaning delete a user, by first stopping it
before removing its data.
There is a new broadcast ACTION_USER_STOPPED sent when a user is
stopped; system services need to handle this like they currently
handle ACTION_PACKAGE_RESTARTED when individual packages are
restarted.
Change-Id: I89adbd7cbaf4a0bb72ea201385f93477f40a4119
Was not properly handling creating DefContainerService when the installation was
for USER_ALL. Not a problem for Market installs.
Bug: 7061571
Change-Id: I4528b4c56e38effa137da56460e78c55b242ba45
This add a new per-user state for an app, indicating whether
it is installed for that user.
All system apps are always installed for all users (we still
use disable to "uninstall" them).
Now when you call into the package manager to install an app,
it will only install the app for that user unless you supply
a flag saying to install for all users. Only being installed
for the user is just the normal install state, but all other
users have marked in their state for that app that it is not
installed.
When you call the package manager APIs for information about
apps, uninstalled apps are treated as really being not visible
(somewhat more-so than disabled apps), unless you use the
GET_UNINSTALLED_PACKAGES flag.
If another user calls to install an app that is already installed,
just not for them, then the normal install process takes place
but in addition that user's installed state is toggled on.
The package manager will not send PACKAGE_ADDED, PACKAGE_REMOVED,
PACKAGE_REPLACED etc broadcasts to users who don't have a package
installed or not being involved in a change in the install state.
There are a few things that are not quite right with this -- for
example if you go through a full install (with a new apk) of an
app for one user who doesn't have it already installed, you will
still get the PACKAGED_REPLACED messages even though this is
technically the first install for your user. I'm not sure how
much of an issue this is.
When you call the existing API to uninstall an app, this toggles
the installed state of the app for that user to be off. Only if
that is the last user user that has the app uinstalled will it
actually be removed from the device. Again there is a new flag
you can pass in to force the app to be uninstalled for all users.
Also fixed issues with cleaning external storage of apps, which
was not dealing with multiple users. We now keep track of cleaning
each user for each package.
Change-Id: I00e66452b149defc08c5e0183fa673f532465ed5
Introduce API to get per-user storage information, keep track
of services associated with users, and various small cleanup.
Change-Id: I5d4e784e7ff3cccfed627d66a090d2f464202634
Get rid of duplication between find/retrieve service funcs; when
a service in a persistent process crashes, restart it immediately
since the persistent process is going to be immediately restarted
anyway; when a new process is attaching, immediately restart any
services associated with it that are waiting to restart, since
it is weird to not let them run if the process comes back for some
other reason.
Change-Id: Id087fe04ebf2b6a4bd00732796c8326364765ea7
Moved a bunch of methods from PackageManager to UserManager.
Fix launching of activities from recents to correct user.
Guest creation APIs
Change-Id: I0733405e6eb2829675665e225c759d6baa2b708f
- You can now use android:singleUser with receivers and providers.
- New API to send ordered broadcasts as a user.
- New Process.myUserHandle() API.
For now I am trying out "user handle" as the name for the numbers
representing users.
Change-Id: I754c713ab172494bb4251bc7a37a17324a2e235e
