Skip to main content

Architecture and Configuration

The purpose of the Memfault Android SDK [source] is to make it really easy to collect detailed diagnostics from your entire Android device, from low-level or hardware issues to high-level application crashes, and upload them automatically to Memfault.

This document provides a detailed reference of the SDK, how it works, and how to configure it.

For a detailed guide on how to integrate the SDK into your devices, see the Getting Started guide.

Collection Types - Bug Reports and Caliper

The Android SDK supports two data collection systems:

  1. Bug reports, a diagnostics reporting tool built into AOSP.
  2. Memfault Caliper, a system designed to precisely specify what diagnostics information to collect on the device for improved performance and privacy.

We do not recommend collecting Bug Reports from production devices, because of the collection overhead, upload size and privacy implications.

info

By default, periodic Bug Report collection is disabled, and all Caliper data collection is enabled. This can be configured in the Memfault dashboard - see Features and Configuration.

Bug ReportsCaliper
Collection mechanismPull: collects all system state, metrics & logs at a point in time. Events/metrics may be missed if they occur long enough ago.Event-driven: collects crashes as they occur using DropBoxManager, periodically collecting logs and metrics.
Performance ImpactCollecting a bugreport (using dumpstate) is CPU-intensive, and can have a noticeable impact (causing stuttering/lag on the device).Caliper collects small chunks of relevant information on a regular basis, with low impact.
SizeBug Reports can be huge files - this is not configurable. Much of what is collected is not useful (e.g. every dumpsys, unfiltered logs).Caliper only collects useful data - so storage/upload sizes are much smaller. This is entirely remotely configurable (collection intervals, log filters, which types of crashes to collect, etc).
PrivacyPII may be contained within a Bug Report (e.g. email addresses in logs, or IP addresses in dumpsys output).Logs can be scrubbed for PII on-device before upload. What data is collected is fully configurable.
LogsLogs are collected at the time of taking the Bug Report. Server-side Log Scrubbing is available.Logs (from all buffers) are collected at a regular interval, and uploaded when a crash occurs. They are scrubbed on-device for PII.
CrashesCrashes are parsed from logs (may be missed if buffers expire).Crashes are processed as they occur, via DropBoxManager. Kernel oops are extracted from regular logcat collection.
RebootsN/ACollected on boot, and visible on timeline/fleet overview pages.
Battery MetricsBattery metrics are collected, and displayed on the Bug Report tab. They are not available as fleet-aggregated metrics.Battery metrics are collected and displayed continuously on the Device Timeline tab. They are also available as fleet-aggregated timeseries metrics.
Installed AppsInstalled app versions can be viewed on the Bug Report tab. They cannot be used as Device Properties/in Device Search.Installed app versions are collected as Device Properties. These can be used for Device Search (or to create Device Sets).
System PropertiesCan be viewed on the Bug Report tab. They cannot be used as Device Properties/in Device Search.Are collected as Device Properties. These can be used for Device Search (or to create Device Sets).
Built-In MetricsN/ACaliper collects device storage/temperature metrics (more metrics will be added here in the future). These can be used as fleet-wide timeseries metrics for charting/device search.
ConfigurabilityWe provide an AOSP patch to optionally enable dumpstate minimal mode - but exactly what this collects is not remotely configurable.Fully configurable on the Memfault dashboard. Settings are synced down to devices, including enabling/disabling each type of data, configuring log filters and scrubbing rules, and tuning collection intervals.

See the Changelog for a complete history of when features were added to the SDK.

Components

At a high level, the process of capturing and uploading diagnostics data involves the following components that work together.

The Bort app

The Bort app is a configurable, updatable app that controls the SDK's behavior and contains the SDK's business logic.

When the Bort app has collected diagnostics data, it will enqueue it for upload, which will take place when the relevant constraints are met.

For security reasons, the Bort app is a privileged app. This means that it must be included in your devices' system image, but it can be updated via over-the-air app updates. While over-the-air updates of the Bort app are not required, they allow for easy SDK behaviour changes and updates to incorporate new functionality.

The UsageReporter app (com.memfault.usagereporter)

The UsageReporter app is a companion system application that provides a mechanism for the Bort app to collect diagnostics information.

The UsageReporter app provides protected APIs for collecting information using two systems, described in detail below.

Memfault Caliper

note

The Caliper system supports trace; per device and fleet wide metrics collection via the Android batterystats subsystem; and log collection via logcat.

Memfault's Caliper system can be used to precisely specify what diagnostics information to collect on the device. We recommend using this system because it allows for:

  • The smallest possible runtime footprint on the device
  • Minimal bandwidth and on-device storage
  • Fine-grained control over what data and when it's collected

Caliper uses the DropBoxManager API to receive traces and crash data from the operating system.

Metrics are collected using the Batterystats subsystem.

Logs are collected from logcat when issues occur. See Logging.

Bug Reports via MemfaultDumpstateRunner

When a bug report is requested by the Bort app, the UsageReporter app invokes the MemfaultDumpstateRunner to capture the bug report.

/system/bin/MemfaultDumpstateRunner

This native program is exposed as an "init.rc service" (see memfault_init.rc for details). This is so that it can be executed in the required dumpstate security context, but get triggered by the Bort app, via the UsageReporter system app. The Bort app runs in the much less capable security context compared to MemfaultDumpstateRunner.

  1. trigger /system/bin/dumpstate (through another init.rc service, memfault_dumpstatez),
  2. copy the bugreport.zip file out of the shell file system sandbox and into Bort's file system sandbox,
  3. broadcast an Intent with the final path such that the Android SDK can process it further.

To permit MemfaultDumpstateRunner to do all the things it needs to do, it is labelled with the existing/builtin dumpstate sepolicy label and is broadened a little further as well (see memfault_dumpstate_runner.te). It's possible to make a tail-made policy that is narrower in scope but this requires more changes to the builtin AOSP system/sepolicy, so we choose to piggy-back on the existing dumpstate type instead. This is also the approach that the AOSP Car product (Android Auto) takes. See sepolicy/file_contexts for reference.

/system/bin/dumpstate

This is the AOSP-built-in Bug Report capturing program; it requires no modification.

The Android SDK provides an optional patch so dumpstate can run in its default behavior, but also in a "minimal" mode. See Minimal Mode below for more information.

Note that it is not triggered through the builtin dumpstatez init.rc service, but through the slightly specialized memfault_dumpstatez. See memfault_init.rc for details on the differences.

Features and Configuration

The Android SDK contains an Over The Air settings system so that settings are defined on the web in your Memfault project, and the SDK will periodically pull the latest settings at run-time. This makes it easy to enable or disable different data sources, or change configuration options. Settings can be updated through the web UI or programmatically via a web endpoint -- no need to deploy an app or OS update.

The settings can be found in the Settings part of the web UI. SDK settings are only visible to and configurable by Project Managers and Organization Admins.

Rate Limiting

Ingestion of data from the Android SDK may be rate-limited. We do not recommend lowering the collection intervals to be less than the default values.

Fallback SDK Settings

As part of the build process, the Android SDK will fetch the current SDK settings and store them as a file called settings.json. We recommend storing this file in your version control system. The Android SDK embeds this file as fallback SDK settings, used temporarily until fresh settings are retrieved from the web, at run-time.

If, during the build process, the Android SDK detects the local settings.json is stale, the build will intentionally fail to alert you that the default settings are out of date. To fix this, remove the stale settings.json file and re-run the build.

If you wish to disable this check, you can skip this step when building the Bort APK by setting a gradle property:

SKIP_DOWNLOAD_SYSTEM_SETTINGS_JSON=1

Bug report capture period

The Android SDK periodically generates bug reports by scheduling a periodic task to trigger the MemfaultDumpstateRunner.

The SDK registers a handler (broadcast receiver) for the system boot event as well as when the app itself is updated or installed. When either of these happens, the app will register the periodic task if one is not registered.

There are configuration options to set the period of this task as well as the initial delay for when the first bug report is generated (e.g. if you wish to wait until more data is available). See SDK settings.

Bug reports can also be triggered programmatically via an Intent.

"Minimal mode" bug report capture

Memfault provides the option to capture "minimal" bug reports. These collect the data required for Memfault diagnostics while being roughly 5x smaller and requiring 10x less load on the system. If system load or bandwidth are concerns for your deployment, we recommend using minimal mode.

If you do not have those constraints and would like to download the bug reports from Memfault to look at deeper diagnostics information, we recommend disabling minimal mode and capturing "full" bug reports.

Enable "minimal" mode via the SDK settings.

Triggering a bug report programmatically

In addition to generating bug reports at regular intervals, you may also wish to capture a bug report if a significant event occurs. Doing this will not affect the scheduled bug reports.

Note that if the dumpstate runner is busy capturing a bug report already, the in-flight bug report will continue and the interrupting request will be ignored.

Triggering a bug report requires that the sender hold the permission specified in the BORT_CONTROL_PERMISSION property in bort.properties. The default is the com.memfault.bort.permission.CONTROL.

An optional ID can be provided via the string extra com.memfault.intent.extra.BUG_REPORT_REQUEST_ID. When provided, the dumpstate.memfault.requestid system property is set to this value. The value can be any string of up to 40 ASCII characters.

Intent("com.memfault.intent.action.REQUEST_BUG_REPORT").apply {
component = ComponentName(
APPLICATION_ID_BORT,
"com.memfault.bort.receivers.ControlReceiver"
)
// Optionally provide a request ID:
putExtra("com.memfault.intent.extra.BUG_REPORT_REQUEST_ID", UUID.randomUUID().toString())

// Optionally enable minimal mode:
putExtra("com.memfault.intent.extra.BUG_REPORT_MINIMAL_MODE_BOOL", true)

// Optionally provide a BroadcastReceiver for status replies:
putExtra("com.memfault.intent.extra.BUG_REPORT_REQUEST_REPLY_RECEIVER",
"com.acme.app/com.acme.app.BortBugReportRequestReplyReceiver")
}.also {
context.sendBroadcast(it)
}

Receiving bug report request status replies

Optionally, a BroadcastReceiver can be implemented to handle the status reply that the Android SDK sends in response to a bug report request. Replies are sent using targeted Intent broadcasts. They are only sent when the com.memfault.intent.extra.BUG_REPORT_REQUEST_REPLY_RECEIVER extra is set to the com.package.id/qualified.class.name of the BroadcastReceiver (see example above).

To register a receiver class called BortBugReportRequestReplyReceiver, add this to your AndroidManifest.xml:

<receiver
android:name=".BortBugReportRequestReplyReceiver"
android:permission="android.permission.DUMP">
<intent-filter>
<action android:name="com.memfault.intent.action.BUG_REPORT_REQUEST_REPLY" />
</intent-filter>
</receiver>

The code to handle the reply Intent would look like something along these lines:

class BortBugReportRequestReplyReceiver : BroadcastReceiver() {
fun onReceive(context: Context?, intent: Intent?) {
if (intent?.action != "com.memfault.intent.action.BUG_REPORT_REQUEST_REPLY") return

// status can be "OK_UPLOAD_COMPLETED", "ERROR_ALREADY_PENDING", "ERROR_TIMEOUT",
// "ERROR_SDK_NOT_ENABLED" or "ERROR_RATE_LIMITED":
val status = intent.getStringExtra("com.memfault.intent.extra.BUG_REPORT_REQUEST_STATUS")

// The request ID that was originally provided:
val requestId = intent.getStringExtra("com.memfault.intent.extra.BUG_REPORT_REQUEST_ID")

Log.i("Got reply for bug report request with ID: $requestId status: $status")
}
}

Triggering a bug report with bort_cli.py

If you wish to generate a bug report using the SDK on a development device over ADB, bort_cli.py provides a convenience command:

./bort_cli.py request-bug-report --bort-app-id your.app.id

Enabling the SDK at Runtime

By default, the SDK assumes it is running on a device that may contain Personally Identifiable Information (PII). As such, it will not run until it is explicitly told to do so (e.g. when user consent has been obtained). Once enabled, that value will be persisted in preferences. If the Bort app's data is cleared, this value must be set again.

To enable the SDK, send an intent to the ControlReceiver. Note that the sender must hold the permission specified in the BORT_CONTROL_PERMISSION property in bort.properties. The default is com.memfault.bort.permission.CONTROL, which may only be may only be granted to applications signed with the same signing key as MemfaultBort (signature) or applications that are installed as privileged apps on the system image (privileged). See Android protectionLevel for further documentation.

Intent("com.memfault.intent.action.BORT_ENABLE").apply {
component = ComponentName(
APPLICATION_ID_BORT, // Whatever you have chosen for the application ID
"com.memfault.bort.receivers.ControlReceiver"
)
putExtra("com.memfault.intent.extra.BORT_ENABLED", true)
}.also {
context.sendBroadcast(it)
}

To disable the SDK, for example if the user later revokes consent, simply send the same intent with the opposite boolean extra

Intent("com.memfault.intent.action.BORT_ENABLE").apply {
component = ComponentName(
APPLICATION_ID_BORT,
"com.memfault.bort.receivers.ControlReceiver"
)
putExtra("com.memfault.intent.extra.BORT_ENABLED", false) // <-- Now disabled
}.also {
context.sendBroadcast(it)
}

If the SDK is running on a device that does not require user consent, this requirement can be disabled by changing a property in bort.properties:

RUNTIME_ENABLE_REQUIRED=false

If you wish to enable the SDK on a development device over ADB, bort_cli.py provides a convenience command:

./bort_cli.py enable-bort --bort-app-id your.app.id

The enable-bort command runs this ADB command under the hood:

adb shell am broadcast --receiver-include-background \
-a com.memfault.intent.action.BORT_ENABLE \
-n your.app.id/com.memfault.bort.receivers.ShellControlReceiver \
--ez com.memfault.intent.extra.BORT_ENABLED true

Setting Project Key at Runtime

A Project Key is required at compilation time of the Bort APK, but it can also be changed at runtime if needed. This can be useful for moving test devices to a separate project so their metrics and issues don't pollute the main project being used. To use this feature, add ALLOW_PROJECT_KEY_CHANGE=true to your bort.properties.

There are two ways to change the Project Key: using a sysprop or using an Intent.

caution

Use either a sysprop or an Intent to change the Project Key, but never both. If a sysprop is configured at build time, then a broadcast cannot be used.

Intent

This is supported from Android SDK 4.5.0 onwards.

Once enabled in bort.properties, you can run the following adb command to change the key.

adb shell am broadcast \
-a com.memfault.intent.action.UPDATE_PROJECT_KEY
-n com.memfault.smartfridge.bort/com.memfault.bort.receivers.ShellControlReceiver
--es com.memfault.intent.extra.PROJECT_KEY "your_new_project_key"

This can be called before the SDK is enabled at runtime, which will ensure that the Project Key compiled into the apk will never be used.

Excluding the com.memfault.intent.extra.PROJECT_KEY parameter will reset to the key compiled into the APK (from Android SDK 4.6.0).

caution

Make sure that the PROJECT_KEY_SYSPROP gradle property is empty, if you intend to change the Project Key using an Intent.

Sysprop

This is supported from Android SDK 4.11.0 onwards.

In addition to ALLOW_PROJECT_KEY_CHANGE=true, a sysprop key must be configured in bort.properties. If your Project Key is defined in the a.b.c.project_key system property, then set PROJECT_KEY_SYSPROP=a.b.c.project_key.

If PROJECT_KEY_SYSPROP is configured, then the Android SDK will check that sysprop every time the main Bort app starts. If the sysprop is set, then this will be used as the Project Key. If the sysprop is not set (or is empty) then the key compiled into the APK will be used.

If the sysprop value is changed after device boot, then be sure to kill the Bort app, so that it will read the new value when it next starts.

note

Changing the Project Key using either mechanism will delete all pending MAR files queued to be uploaded.

Software Versions, Hardware Versions and Device Serials

It is possible to customize which values to use as the Software Version, Hardware Version or Device Serial for a given project. For background on software and Hardware Versions, see Software and Hardware Versions.

The customization is done during the project creation. To change these settings, you must create a new project.

The Android SDK automatically retrieves these settings during the APK build process so no additional configuration is required.

Install the SDK to an alternate location

The default SDK location is in vendor/memfault/bort. If you would prefer to place the SDK at a different location in your source tree, first clone the SDK to the desired location in the source tree, then find and replace instances of vendor/memfault/bort with your target directory.

For example, to place the SDK in the system partition, clone the SDK to packages/apps/bort then run find and replace (e.g. on BSD/macOS):

LC_ALL=c find . \( -type d -name .git -prune \) -o -type f -exec sed -i '' "s/vendor\/memfault\/bort/packages\/apps\/bort/g" {} +

Data Reporting Frequency

There are two modes in which the Bort app can be configured to send data. The first is by bundling all data into a MAR (Memfault Archive) file, or by sending each piece of data individually to Memfault as it's collected from the system (deprecated - see below).

note

During development, you will want to increase the frequency at which data is sent. To do this, enable Developer Mode

Uploading MAR files

When uploading data via MAR files, the Bort app collects diagnostic data for 2 hours and then bundles it all together in a single MAR file and uploads it.

Depending on the time at which data was collected from the system, it may take up to 2 hours for it to be sent to Memfault.

info

The use of MAR files is required to use Memfault's Fleet Sampling feature.

Uploading individual files

info

This method of uploading files from the Android SDK is deprecated as of Android SDK 4.1, and was completely removed in Android SDK 4.6.0.

This configuration would upload individual files to Memfault. After the data is collected from the system, it is uploaded up to 15 minutes later.

Sync Behavior and Triggers

Memfault uses the Android JobScheduler APIs to schedule its work following Android's best practices. To modify this behavior for your project see the following options.

Reducing Data on Metered Networks

Two Project settings control if data sent when on a metered network. The "Allow Uploads on Metered Networks" and "Allow OTA downloads on Metered Networks" settings available in your Project Settings - Data Sources configure Memfault to put either a NetworkType.CONNECTED (if true) or NetworkType.UNMETERED constraint on its network requests.

Sync Rates

Memfault splits sync data into two different categories, OTA releases and data contained in Memfault Archives (MAR). MAR data consists of metrics, logs, and DropBoxManager Entries collected by Memfault.

TypeDefault Rate
MAR Data2 hours
OTA Releases12 hours

These settings can be adjusted if desired by contacting support.

Retries

OTA requests will retry linearly at 15m intervals (15m, 30m, 45m, ..).

All other requests will retry exponentially starting at 5m (5m, 10m, 20m, 40m, ..).