Analytics

Metrics

Metrics are declared in include/pbl/analytics/analytics.def as unsigned, signed, scaled, timer or string metrics, and updated by name from anywhere, apps included, since the functions behind the macros are syscalls:

PBL_ANALYTICS_ADD(button_pressed_count, 1);

PBL_ANALYTICS_TIMER_START(backlight_on_time_ms);
PBL_ANALYTICS_TIMER_STOP(backlight_on_time_ms);

Collectors

A metric that is only worth computing once per period comes from a collector: a function next to the state it reads, called on the system task right before each heartbeat. Collectors are gathered at link time, so analytics does not know about them and a firmware without the code has no collector to stub:

static void prv_collect_pfs_stats(void) {
  PBL_ANALYTICS_SET_UNSIGNED(pfs_space_free_kb, get_available_pfs_space() / 1024);
}
PBL_ANALYTICS_COLLECTOR_DEFINE(prv_collect_pfs_stats);

They run in no particular order, so a collector must not depend on another.

Heartbeats

The native backend, subsys/analytics/native.c, keeps the current values. Each heartbeat it writes them as one packed record to a data logging session, which sends it to the phone, and resets them for the next period. Running timers keep running across heartbeats.

The record’s layout follows analytics.def, and it carries the firmware’s build ID. tools/analytics_heartbeat_layout.py extracts the layout from the matching ELF file, for parsers.

Shell