Test: added e2e tests for count/duration metrics sliced by fields in condition and with/without links.
Change-Id: Ie34deba68e6780abdde458be3f0ce5284e76a1a2
Records start and duration of shutdown, along with reason and if it was
a reboot.
Test: manually verified statsd received atom. CTS test for this will be
difficult, and I will investigate further later.
Change-Id: I0f6b595e0e251fd0a8b38127182d055885460a55
Defines following atoms:
-- AppStartMemoryStateChanged
-- ProcessMemoryState
-- LmkStateChanged
-- LmkProcessKilled
These changes are based on the discussion with WestWorld team.
See: go/android-p-memory-metrics
Bug: 72177881
Test: Manual
Change-Id: I884f23f3d7627440af5faa25f9479ba0d8763bec
1/ Avoid unnecessary field/dimension proto construction.
2/ use unordered_map for slicing.
3/ Use dimension fields to compare dimension keys.
Test: all statsd tests passed.
Change-Id: I2f74f78589b7f6ecd0803a2ead822b8d0399f334
+ refactor pullers in StatsCompanionService to be more modular
+ rename CpuSuspendTime and CpuIdleTime to SystemElapsedRealtime
and SystemUptime
Test: will add cts test
Change-Id: I463103fb271511cef4e0f877c20fd167fe8b173b
+ Bugreport will use the non-verbose mode
+ Reuse the log_msg object in LogReader
+ Add logd errors to StatsdStats
Bug: 72383073
Test: manual + statsd_test
Change-Id: Id5a8b103074d034f5ece3c9831c740d44a5df9cd
Created an enums.proto that contains the device idle modes,
which is referenced by batterystats.
Note that, currently, batterystats is the origin of these
constants in the java code, and I have kept that as is.
Thus, batterystats references the enum.
Nevertheless, because the actual control of device idle
mode is DeviceIdleController.java, where these constants
will likely one day be moved, I have put the constants
in server/enums.proto (since DeviceIdleController is
in server, not os like BatteryStats).
Bug: 69478930
Test: cts-tradefed run cts-dev -m CtsStatsdHostTestCases -t android.cts.statsd.HostAtomTests#testDeviceIdleModeStateChangedAtom
Change-Id: I6e66801ae8256aab423067f9a4b852a194564a8d
App transitions, from ActivityManagerInternal, are now referenced via a
proto enum. This is also referenced by statsd's atoms.proto.
Bug: 69478930
Test: still compiles
Change-Id: Ifb5cb0120d4cd4e2464ce3b5a7455842cb55259c
For frameworks constants that don't have intrinsic meaning (i.e. their actual
value and order don't matter), so that it is unlikely that their values
will be changed:
This cl introduces proto enums representing some constants found in
the Android codebase, and connects the two.
By using the Proto enum as the source-of-truth, it means that Java and
proto can be kept in sync. Otherwise, when the Java frameworks code
changes, it silently breaks the protos from working properly, since the enums
are wrong. By having the Java code reference the proto enums, it ensures
that everything is in sync. The values of the constants are unchanged.
But future changes to these constants will need to be done in the proto
file, which the Java file merely references.
The protos are necessary for incidentd and statsd and, in the future,
possibly dumpsys. In this way, the logging mechanism is much less likely
to get broken when new constants are added, and we can be ensured that
the logging accurately reflects the underlying codebase.
Bug: 69478930
Test: cts-tradefed run cts-dev -m CtsStatsdHostTestCases
Test: cts-tradefed run cts-dev -m CtsIncidentHostTestCases
Change-Id: If79032c34b2799db1e3e70cb47b1312fd72092b9
Allows a uid that uploads a statsd config to additionally
register a BroadcastSubscriber with statsd. If statsd
detects an anomaly (according to the config's Alert),
statsd can inform a BroadcastSubscriber provided in the config.
The config uses a subscriberId (just an int) to identify the
BroadcastSubscriber. It then uses StatsManager.setBroadcastSubscriber
to associate that subscriberId with a given PendingIntent.
Then, when the anomaly is detected, statsd sends a broadcast
using that PendingIntent, alerting whoever was specified by
the config/setBroadcastSubscriber.
Bug: 70356901
Test: cts-tradefed run cts-dev -m CtsStatsdHostTestCases -t android.cts.statsd.alert.BroadcastSubscriberTests
Change-Id: I4d9ea9a6c8a85e61fadfd99c1513c55abbadd5e9
Rename the activitymanager.proto to enums.proto, in accordance with the
new convention on naming these protos.
Also, changes processStateAmToProto from simplying multiplying by 100 to
a switch statement, to be more future-proof.
Also, changes the values of the proto's process state, to reflect that
the values will not be kept in sync with ActivityManager.
Note that this cl was originally started in ag/3008434, and partially fulfilled in ag/3114337.
Bug: 69478930
Test: run cts-dev -m CtsStatsdHostTestCases -t android.cts.statsd.ProcStateAtomTests
Change-Id: I2b3bf2552b879c74d8985338df5a57c01850cb91
Turns out that statsd leaves some file descriptors opened
without the O_CLOEXEC flag. This CL mass-closes file descriptors
up to a realistic number of FDs. This is to avoid SELinux complaining
about perfetto accessing these files from the statsd domain.
With this change perfetto works with statsd without disabling SELinux.
Relevant SELinux CL:
https://android-review.googlesource.com/c/platform/system/sepolicy/+/598774
Change-Id: I745d621937fbc9b20a4c733948cd0dd24dd6e8b2
- Limit total number of files to 1000
- Limit total size of files to 5MB
- Remove idle files to be deleted after 30 days
Bug: 69854160
Test: manual testing, statsd, statsd_test
Change-Id: I33148a3b7ca11d413ec2495d5c0659f1ba4485c3
+ We should benchmark the core functions in the critical path.
1. LogEvent parsing
2. Log matching
3. Event processing in various metrics
....
+ Using microbenchmark, we can quantify the optimization we do in the future.
---------------------------------------------------------
Benchmark Time CPU Iterations
---------------------------------------------------------
BM_LogEventCreation 4772 ns 4705 ns 145738
Test: make -j64 statsd_benchmark
Bug: 72242322
Change-Id: I43aa704ffdc7cd21e02ef9038eff66ca7022dbfb
Statsd will contain at least one snapshot of the uid map. The
previous design was not very robust in case a snapshot was missing.
Also fixes subtle bug with updating the isolated uid mapping since
this should always be kept up to date even if there are no metrics
being used (since metrics may be added later after the isolated uid
was created).
Test: Checked that unit-tests pass on marlin-eng.
Change-Id: I99754ed9016d980564e409b0946a46b398fd12b7
