Commit Graph

47 Commits

Author SHA1 Message Date
Yangster-mac
d40053eb8b Map isolated uid to host uid when processing log event in statsD.
Test: added test case for isolated uid in Attribution e2e test.
Change-Id: I63d16ebee3e611b1ef0c910e5154cf27766cb330
2018-01-09 21:45:46 -08:00
Yangster-mac
b0d0628a29 Thread-safety at log processor level.
Test: statsd unit test passed.

Change-Id: Ibe8c8d3cc8297875b16ee385c077b71c87353147
2018-01-08 14:59:42 -08:00
Yao Chen
147ce60278 use only string type in the log source whitelist.
+ predefined "AID_X" will be provided as string type to statsd, and we will translate
  to integer uid using the static map.

Test: statsd_test

Change-Id: Ie47d8481e0c456457e6881ebb9cb4ce008e772b8
2018-01-04 09:57:03 -08:00
Yangster-mac
94e197cceb 1/ Change all "name" to id in statsD.
2/ Handle Subscription for alert.
3/ Support no_report_metric

Bug: 69522276
Test: all statsd unit tests passed.
Change-Id: I851b235f2d149b8602b0cad632d5bf541962f40a
2018-01-03 15:34:00 -08:00
Yangster-mac
2087716f2b 1/ Support nested message and repeated fields in statsd.
2/ Filter gauge fields by FieldMatcher.
3/ Wire up wakelock attribution chain.
4/ e2e test: wakelock duration metric with aggregated predicate dimensions.
5/ e2e test: count metric with multiple metric condition links for 2 predicates and 1 non-sliced predicate.

Test: statsd unit test passed.

Change-Id: I89db31cb068184a54e0a892fad710966d3127bc9
2018-01-01 10:01:36 -08:00
Bookatz
857aaa5208 Splits AnomalyTracker into two files
Splits out DurationAnomalyTracker-specific functions into their own
subclass.

Test: the unit tests and CTS tests
Change-Id: Id6eb74d232b4a9c3a932d805d1ba3f0ba43a88b1
2017-12-22 15:05:31 -08:00
Yao Chen
d10f7b1c7b Add log source filtering in statsd to filter out spams.
+ Add log source whitelist in StatsdConfig
+ Some changes in UidMap API. Listener needs to be wp instead of sp.
+ Update dogfood app config to have log source
+ Increase the stats service thread pool size to 10 (9+1).

TODO: add unit tests(b/70805664). This unit test takes some time to write.

Test: statsd_test & manual

Change-Id: I129b1cc13db5114db7417580962bd7cc4438519d
2017-12-20 18:45:43 -08:00
Chenjie Yu
85ed838713 align metrics to 5min bundary
We use one alarm clock for all pulled atoms.
If metrics from different configs are not aligned,
the clock will be set to repeat at higher and higher
frequency, and consume a lot of battery.
Current implementation assumes a 5min minimum bucket
size. New metric start time is set to be aligned to
the start time of statsd in the next 5min.
So it will ignore events up to 5min.

align puller alarm to minute bundary

Test: unit test
Change-Id: I77ffa3c13de363c780b1000181b9a9b780dd0846
2017-12-16 17:08:46 -08:00
TreeHugger Robot
2d48f13c7d Merge "Only create ProtoOutputStream when onGetData() is called." 2017-12-13 23:16:59 +00:00
Yao Chen
288c600013 Only create ProtoOutputStream when onGetData() is called.
The exception is EventMetricProducer. Each EventMetricProducer will still have a ProtoOutputStream
Because LogEvent comes as a fixed 4K, it's more memory efficient to have an 8k ProtoOutputStream for
storing the events.

Also removed finish() api in MetricProducer, which was intended to use with Dropbox.

Test: statsd_test & dogfood app
Bug: 70393808
Change-Id: I2efe4ecc76a88060a9aa5eb49d1fa6ea60bc5da8
2017-12-12 14:20:32 -08:00
yro
03faf09330 Migrate disk directory from /data/system/ to /data/misc/
Test: statsd, statsd_test
Change-Id: I6d2fe97afd79fb9b36d180d5e6e6a7a166a228b7
2017-12-12 00:17:50 -08:00
David Chen
d9269e2ee7 Adds rate limit to checking byte size.
Since there is a separate guardrail for memory used by uid map, we
no longer add the memory from uid map with the memory per each
config's metrics. We also prevent the byte size check from happening
too frequently. In order to mock the MetricsManager, we refactor
some of the existing methods.

Test: Added unit-tests and verified they all pass on marlin.
Change-Id: I15cf105f7d95f4016fdb0443b0a33eebe862cafb
2017-12-07 18:22:58 -08:00
Yao Chen
312e898325 Let the event flow to MetricsManager. Comment out the size check only.
Test: statsd_test, and manual
Change-Id: I862967510eaf4d402471e9dd6b9c85f6037dd7e1
2017-12-05 15:29:03 -08:00
Yao Chen
0b73ccad8f Urgent fix. Once UidMap size exceeds the limit, statsd triggers data drop every time a log comes in
Test: manual
Change-Id: Idf93e5aca19b80acf964670fa4bc9f1f0781df1f
2017-12-05 09:37:06 -08:00
David Chen
12942956bc Tiny fix to bug when statsd should clear data.
Previously, we always sent a broadcast, even after we have exceeded
the memory limit for this config key. We switch the order so that
we drop the data if the limit is exceeded. If greater than 90% of the
way to the limit, we send the broadcast.

We need to find a way to unit-test this behavior.

Test: N/A.
Change-Id: I6ea40b9e34dceb19805d9af24495d72878f787e0
2017-12-04 14:28:43 -08:00
David Chen
c136f45aee Adds guardrail for memory usage for statsd uid map.
Checks if current memory usage of uid map is above a configured limit
and if so, we start deleting snapshots. If there are no more
snapshots, we begin deleting two of the deltas. Also records stats
in the guardrail StatsdStats. Also fixes an edge case where a config
is added after the snapshots are added. We request a snapshot of all
installed uid's at that moment. Finally, adds the uid map memory size
when determining if we should send a broadcast to trigger collection.

Test: Added unit-tests and check they pass on marlin.
Change-Id: Id5d86378bd1efe12a06b409164c777c0c6f4e3ab
2017-11-29 14:55:15 -08:00
TreeHugger Robot
3e585ecb51 Merge "Some fixes in StatsdStats, and add some unit tests" 2017-11-28 06:58:54 +00:00
Yao Chen
69f1baf7dd Some fixes in StatsdStats, and add some unit tests
+ Add timestamp for when metric data is reported.

Test: statsd_test

Change-Id: Ief5ec5172feed4ec74b7422b77cf69ec8361ef2f
2017-11-27 20:45:16 -08:00
Bookatz
cc5adef2d0 Statsd anomaly detection - fixes
Fixes a few items in AnomalyTracker, especially to do with what happens
when an anomaly alarm fires.

Test: unit tests still pass
Change-Id: Ia89bd617442e952e587336b890c3ca67430b5e21
2017-11-27 15:35:40 -08:00
Yao Chen
b356151e63 Add StatsdStats and guardrail.
+ StatsdStats is the global class that tracks the stats about statsd.

+ Added guardrail for classes that have a map which could potentially grow
  unboundedly with the number of logs.

TODO: add unit tests & CTS for StatsdStats, and guardrail
      add stats for pulled atoms.

Test: statsd_test

Change-Id: I0ea562de4dd3f6162f7923a9c193420b482c1d51
2017-11-27 10:52:54 -08:00
yro
947fbce521 Captures metrics on disk when devices reboot and shutdown. Specifically,
1. Create intent receiver in StatsCompanionService to listen to shutdown
events.
2. Create StatsWriter class to handle disk writes and deleting files.
3. Update StatsLogProcessor, ConfigManager, and StatsService to handle
files on disk using StatsWriter.
4. Add a wrapper for ConfigMetricsReport.

Still TODO is to be able to add a guardrail to prevent accumulating
excessive amount files on disk, which will be followed up by another
change.

Test: statsd, statsd_test
Change-Id: Ia0b3af315af545daa8b0078b3700c600aa7c285f
2017-11-22 18:39:23 -08:00
Yangster
7c334a129e Make member function as const whenever possible.
Test: unit tests passed.
Change-Id: I751cabf305a4b5aa2095853cc951837da0df4c78
2017-11-22 14:28:00 -08:00
Yangster-mac
e2cd6d509b 1/ Duration anomaly tracker with alarm.
2/ Init anomaly from config based on the public language.
3/ Unit tests for anomaly detection in count/gauge producer.
4/ Revisit the duration tracker logic.

Test: unit test passed.
Change-Id: I2423c0e0f05b1e37626954de9e749303423963f2
2017-11-20 15:37:24 -08:00
David Chen
1d7b0cd6b5 Support StatsD sending broadcasts.
StatsD will send a broadcast when we're 90% of the way to our
allocated memory limit for the configuration. If the memory usage
goes over the limit, we just lose all the data for this config.

Also modifies the adb shell commands to facilitate debugging of the
broadcasts.

Test: Manually tested on marlin-eng with custom gmscore code.

Change-Id: I517a15bd4c959aa221802f84a51f13141a725102
2017-11-17 14:57:50 -08:00
Stefan Lafon
ae2df01aae Rename "stats_events" as "atom".
Also make atom WakelockStateChanged use the "Type" enum instead of int32.

Test: Unittests are passing. Statsd is working.

Change-Id: I0909e2d97297f78996a81366d66aae62d5bf5ce1
2017-11-14 11:56:05 -08:00
yro
b0378b093d 1. Add FIELD_COUNT_REPEATED bit masks to repeated fields that were previously missing
2. Resolve segfault when serializing proto into vector of uint8_t
3. Add a local byte size variable for EventMetric data as we cannot call
size() of ProtoOutputStream before we finish writing.
4. Replace hardcoded conversion of second to nano second with NS_PER_SEC

Test: statsd, statsd_test

Change-Id: I47a2ce2b05e6191c18596489682118edcb41e945
2017-11-10 11:09:20 -08:00
TreeHugger Robot
570d4db0c9 Merge "Tracks isolated uid's and their parent uid." 2017-11-09 23:53:58 +00:00
David Chen
21582961d7 Tracks isolated uid's and their parent uid.
We push events from BatteryStatsImpl if an isolated uid is added or
removed and we have a custom rule in statsd to use these events to
update our uid map. In the future, we need to use this map to
convert all incoming uid's to their host uid.

Test: Added unit-test to UidMap_test.
Change-Id: I33c0451eb2c886161f22dd12e479d216fad0940d
2017-11-09 13:39:43 -08:00
yro
17adac9cf3 Finish migrating to use ProtoOutputStream. This change will take the
outputs of each MetricProducer's and merge the results into
ConfigMetricsReport which holds ConfigKey, repeated field of
StatsLogReport, and UidMap. The data will be represented as
vector<uint8_t> which can be passed down to binder call for clients to
pick up. Also, all unnecessary dependencies to stats_log proto have been
removed.

Test: statsd, statsd_test
Change-Id: Ia69137cbc8613644a892e6be1e87b4858bd39fe3
2017-11-08 23:40:28 -08:00
Yao Chen
93fe3a34a0 Add unit tests for CountMetricProducer, EventMetricProducer
And other miscellaneous fixes.
+ clang-format
+ 2 bug fixes, one in dump-report command, one in ResourcePowerManagerPuller

Test: statsd_test

Change-Id: Ibd164d948ad62adcc529d813df1210781e38be47
2017-11-06 17:20:10 -08:00
David Chen
d689689891 Updates uidmap to update snapshots and upload.
We send a snapshot of all installed apps with their uids every time
a user is added or removed and when statsd is started.
We keep track of the latest timestamp when a config key has retrieved
the UID map data. This allows us to remove older data when we're
guaranteed that all config sources have retrieved the old data.

Test: Added more unit tests to UidMap_test and passed on marlin-eng.
Change-Id: I34a3d61e75eedec44b98d896d7f6db0bc383f46a
2017-10-31 16:20:53 -07:00
yro
69007c8262 Implement a logic to rate limitted flush statsd with by checking the
size of each MetricsProducer's. The implementation of byteSize() method
is still TBD as it depends on migration to ProtoOutputStream.

Test: statsd, statsd_test
Change-Id: I966606044d7cb814dabe94192bacecad91f28177
2017-10-30 21:01:08 -07:00
yro
fd05a4ee56 Remove dropbox dependency from StatsLogProcessor
Test: statsd, statsd_test

Change-Id: Ia9a3e9b2b3ffffaed81f6045ec43b7d908947fb6
2017-10-26 14:18:25 -07:00
yro
31eb67b349 Adds aidl definitions and their implementations for binder transfer of
statsd entries to clients. This change only includes changes on statds
side and does not include java library for clients to import. Java
library will be a separate change as it requires system api review.

Test: statsd, statsd_test
Change-Id: I306c6e9687801668cc0145b12d38406bfe634775
2017-10-24 15:25:27 -07:00
Yao Chen
729093df0d Add support for dimension, and link with condition and added DurationMetric
Now we support following metrics:
  <Duration> of [app holding a wake lock], while [*this app*] is [in background] [AND] [screen is off]
  [Slice] the output by [app name, wake lock name], with bucket size [30sec]

+ Also added onDumpReport() api in MetricsManager, it can be called from client to fetch the data from
  statsd

+ Also added command line tool to dump the StatsLogReport from all metrics for debugging.

+ Synced proto from google3. with a pending cl (cr/172359050)

TODO: We need to add tons of tests to test the Metrics. I will work on it after this CL so people
can be unblocked.

I locally test the duration metric with wake lock with an app that generates StatsLog events.

Test: statsd_test

      and manual test, and run:

      adb shell cmd stats dump-report
      We have a default config, which contains a metrics to count PROCESS_START event sliced by
      package name.

Change-Id: I4838cc6cf025c143b7e84f43040703a78121fd25
2017-10-19 15:40:55 -07:00
Joe Onorato
c4dfae56c1 Introduce a new wrapper for log_msg -- LogEvent
It stores all of the parsed fields in a single vector, and
provides accessor methods to get at fields by index with
their correct type.

Test: statsd_test
Change-Id: I4fa94e4ce52db3ac87f19b62f9c85398de6e8145
2017-10-18 04:19:04 -07:00
Joe Onorato
9fc9edf95a Clean up how we handle configurations, and other assorted cleanup
- Add a ConfigManager class that tracks the configurations
  that have been passed to us.  Configurations are now
  tracked by tuples of (uid,tag), where the tag is an
  app-defined string, in case a single uid has multiple
  configurations.
- Move all of the initialization into StatsService.
- Get rid of the ability to have multiple LogListeners. Raw
  events are now pushed directly into StatsService, which
  can distribute them to the interested parties (and will
  eventually be able to do the proper locking).
- Add Log.h, which sets our LOG_TAG correctly.
- Move some of the related files that I expect will grow some
  into their own subdirectories.

Test: statsd_test
Test: adb shell cmd stats config ...
Test: adb shell dumpsys stats
Change-Id: I79487603003d8a842d5bd319741f1ecbf72063d1
2017-10-18 04:18:58 -07:00
David Chen
de70169109 UID mapping to provide app name and version.
The UID map is updated by StatsCompanionService, which listens to broadcast
updates indicating that an app was updated/installed or removed. Also,
the entire map is updated when statsd first connects to the companion
service. Also, there is a way for metrics producers to subscribe to
updates, so that they can know when an app was upgraded.

Test: Created new unit-test for mapping and manually tested for install
and remove. Did not manually test the app upgrade.

Change-Id: I6676ae5c93b75c72d9badabb36aa9c40006db07d
2017-10-10 14:33:23 -07:00
Yao Chen
caf339d004 More complete implementation for condition and log matchers in statsd.
+ also synced proto from google3 to fix the LogEntryMatcher proto

+ MetricsManager represents StatsdConfig, it's responsible for initializing and managing all
LogEntryMatcher, Condition, and Metrics. Start review from here.

+ Added more complete StatsdConfig initialization, including building the map for:
    LogEntryMatcher -> Metrics
    LogEntryMatcher -> Condition
    Condition       -> Metrics.

    All the maps use index(int). The extra amount of memory for storing mappings help us
    quickly process log events.

  The StatsdConfig initialization process detects malformed config
  - Circle dependency
  - Missing definition
  etc.

 And once we detect ANY error, statsd will reject the config. And the resources related to this
 config will be released.

Test: Added unit tests
Change-Id: I2c4aefdbf3e2aa1701eacbb2fb5e653819ec1fbb
2017-10-10 10:03:46 -07:00
Yao Chen
44cf27c148 Add metric computation skeleton to statsd.
This cl is to let statsd understand statsd_config, and compute metrics
 defined in the config.

+ StatsLogProcessor is given a StatsdConfig (hard coded right now).
  We construct a MetricProducer for each of the metric, and the metrics
  share Condition and LogEntryMatchers

+ Added the CountMetricProducer type for CountMetric.

We can now count times of SCREEN_ON events given a config.

TODO: 1) conditions are not implemented.
      2) slicings are not implemented in CountMetric
      3) move the interaction to dropbox to a separate thread
      4) decide how the in memory metrics would be used by anomaly detection

Test: manual test.

      $ adb shell /system/bin/statsd

      $ cat config_file.dat | adb shell cmd stats config

Change-Id: I38f4059c0dc5a827c338131d4a6fa7d4cbe865db
2017-09-29 12:52:41 -07:00
yro
fbeb4cb3af Merge multiple eventMetricData entries into a single StatsLogReport
when writing to dropbox

Test: manual test.

Change-Id: I8ce323002ae1cad62703ece92bd99a5058b16fba
2017-09-24 22:42:44 -07:00
Yao Chen
3f0c1f2368 Fix a timestamp bug.
Test: manual
Change-Id: I8067fa8029c0b5702cae09202f8f9f7e443e902d
2017-09-22 21:25:11 -07:00
Yao Chen
ef99c4fa23 clang-format existing code in statsd.
Added .clang-format, because there isn't an official .clang-format file for Android framework code.

before we upload changes, do:

clang-format -style=file -i [file list]

to format the files that you touched.

Test: formatting only. NO code changes.
Change-Id: I90e87f1ee6618da8ea9bc2221c609c415a4046a8
2017-09-22 16:26:54 -07:00
Bookatz
906a35c814 Statsd namespace is defined and used
Statsd code now lives in android::os::statsd namespace. Existing files
are largely modified to follow this convention.

Exception: parse_util, since it seems a bit different.

Test: code compiles and existing statsd_tests still pass
Change-Id: Idf92a071b9ed172d01eb3087a4fa3609d67a038c
2017-09-20 16:58:14 -07:00
yro
00698daf3c Translate Android log entries to stats_log proto
Test: tested on local device

Change-Id: If9a779a96d31cefaffb1e4424629c14b08e1fc57
2017-09-18 23:30:14 -07:00
David Chen
0656b7a158 Adding ability to add configs via adb command-line. The input
must be in serialized binary format. Also fixes small issue
in build rule related to proto lib.

Test: Manually tested that ADB command works without crashing.

Change-Id: Iba2e677561ff500adb601a598f73e8a7b32540e5
2017-09-15 14:07:52 -07:00
Yao Chen
ab273e2e13 Add a DropboxWriter in statsd.
+ The DropboxWriter keeps data in cache, and flush to files once the
  size exceeds the maximum value.

+ Different components should create their owner DropboxWriter with
  different tags, e.g., anomly detection, experiment metrics, etc.

+ Copied stats_log related protos from g3

Test: run statsd, and adb shell dumpsys dropbox
      Will add unit tests.

Change-Id: If06e9a9953be32082252b340a97124d732656b40
2017-09-13 16:58:40 -07:00