v5.4.0 - July 1, 2025

🚀 New Features

• Added 3 more logs-to-metrics parser types.
  • sum_matching captures 1 integer, and sums all captured values in the
    heartbeat.
  • distribution_minmeanmax captures 1 double, and generates a min/mean/max
    distribution for all captured values in the heartbeat.
  • string_property captures 1 string, and reports the latest value for that
    string.
• Added tracking of the metered, temporarily unmetered, and roaming network
  capabilities in HRT.
  • Added a connectivity.metered.latest heartbeat metric for tracking metered
    network usage.
• Added new disk wear and disk write Device Vital heartbeat metrics. eMMC/UFS
  lifetime estimation and version stats are collected from the Health HAL
  implementation, falling back to a hardcoded UFS or eMMC filepath if not
  implemented. Bytes written each heartbeat is collected by reading the sectors
  written value from /proc/diskstats for matching physical devices in
  /sys/block multiplied by the sector size to get a value in bytes.
  • disk_wear.<source>.lifetime_remaining_pct.latest captures the lifetime
    remaining estimation for type A flash in increments of 10%. The source can
    be one of: "mmc0", "624000.ufshc", or "HealthHAL".
  • disk_wear.<source>.lifetime_b_remaining_pct.latest captures the lifetime
    remaining estimation for type B flash in increments of 10%. The source can
    be one of: "mmc0", "624000.ufshc", or "HealthHAL".
  • disk_wear.pre_eol.latest captures the pre-EOL status of consumed reserved
    blocks, as "Normal", "Warning", or "Urgent".
  • disk_wear.version.latest captures the version as described from the vendor
    implementation of the Health HAL.
  • disk_wear.vdc.bytes_written captures the bytes written for that heartbeat.
• Added per-app CPU usage metrics. The CPU usage percentage for each heartbeat
  is collected by parsing /proc/<pid>/stat/ for system and user time
  (stime + utime) divided by the total usage (sum of all fields from
  /proc/stat).
  • By default, the SDK will only create heartbeat metrics for a defined set of
    "interesting" processes. Please reach out to customer support to configure
    this set.
  • Otherwise, the top 10 processes exceeding the 50% CPU usage threshold will
    also be recorded. Please reach out to customer support to configure these
    thresholds.
• Added support for Android 15.

📈 Improvements

• Added parsers for more HRT batterystats data.
  • screen_doze captures whether the display is dozing in a low power state
    (link).
  • flashlight captures whether the flashlight was turned on.
  • bluetooth captures whether Bluetooth was enabled.
  • usb_data captures whether a USB connection was established (as reported by
    android.hardware.usb.action.USB_STATE).
  • cellular_high_tx_power captures whether the modem spent more of its time
    at the highest power level versus any other level.
  • nr_state captures the service level of the 5G network that's connected
    (link).
• Updated comments on various batterystats fields. More improvements to come in
  clarifying the name and use of batterystats metrics!
• Dumped any existing logs when continuous logging starts, to retain logs
  captured from before a reboot.
• Added control over the maximum age of sampled data, to match the controls over
  unsampled data.
• Added support for parsing the binary log buffers (events, security, stats)
  with continuous logs.
• Modified dumpstate to write bugreports to /data/misc/MemfaultBugReports
  instead of to Bort's app dir. This is necessary for Android 15 support and CTS
  compliance to work around neverallow sepolicy rules. This also allowed
  deleting the custom bort_app_data_file app label for the Bort app.

🚧 Fixes

• Added explicit Android System AID mappings for more consistent system UID
  attribution in tombstone and network stats parsing.
• Added VPN and USB as explicitly defined network types using
  connectivity.network.
• Fixed a bug in PhoneState where the unknown ??? state was reported as
  null.
• Added exponential backoff for EAGAIN errors in continuous logs. Each write is
  fsync'd immediately after write to reduce the likelihood of data loss.
• Added missing ignore.cil files for SDKs 32, 33, and 34.

🏠 Internal

• Stopped running unit tests on release sources.
• Migrated the SDK to communicate over a named domain (*.memfault.test) rather
  than localhost for local development.
• Removed unneeded mockk init calls.
• Added some tests for unknown fields in device config.
• Refactored the unsampled holding area logic into a single class.
• Differentiated some continuous logging logs for easier debugging.