Rongsong Shen
2e8ed8cb4e
Add operational metrics infrastructure for the XCE controller persistence and HA subsystems (design doc P3-7, review A8-007). - src/include/ctrl_metrics.h: declares 7 _Atomic uint64_t metrics with ctrl_ prefix, CTRL_METRIC_INC / CTRL_METRIC_SET macros, ctrl_metric_desc_t descriptor struct, ctrl_metric_descriptors[] table, and ctrl_metrics_init() - src/controller/ctrl_metrics.c: defines the 7 counters/gauges, the descriptor table (7 entries + NULL sentinel), and ctrl_metrics_init() - src/controller/persist.c: #include ctrl_metrics.h; call ctrl_metrics_init() at persist_start() entry; CTRL_METRIC_INC(ctrl_repl_degraded_count) when repl_wait_ack returns < 0 (degraded / local-only durability mode) - src/controller/meson.build: add ctrl_metrics.c to xce_controller and controller_unit_test source lists; add test_ctrl_metrics.c to test sources - src/controller/testsuite/test_ctrl_metrics.c: 5 TDD-first Unity tests covering init-zeros, increment, set/overwrite gauge, descriptor sentinel count, and transient-stuck counter accumulation — all PASS
68 lines
2.9 KiB
C
68 lines
2.9 KiB
C
/*
|
|
* ctrl_metrics.c — P3-7: Operational metrics for the XCE controller.
|
|
*
|
|
* Defines the 7 atomic counters/gauges and the descriptor table used
|
|
* for metric discovery. All variables are initialised to 0 at program
|
|
* start (C guarantees this for file-scope variables), and are zeroed
|
|
* explicitly by ctrl_metrics_init() which is called from persist_start()
|
|
* so that tests can call it repeatedly without interference.
|
|
*/
|
|
#include "ctrl_metrics.h"
|
|
|
|
#include <stdatomic.h>
|
|
#include <stddef.h>
|
|
#include <stdint.h>
|
|
|
|
/* ── Counter/gauge definitions ──────────────────────────────────────── */
|
|
_Atomic uint64_t ctrl_repl_degraded_count = 0;
|
|
_Atomic uint64_t ctrl_split_brain_detected = 0;
|
|
_Atomic uint64_t ctrl_async_events_lost = 0;
|
|
_Atomic uint64_t ctrl_transient_jobs_stuck = 0;
|
|
_Atomic uint64_t ctrl_wal_bytes = 0;
|
|
_Atomic uint64_t ctrl_checkpoint_duration_ms = 0;
|
|
_Atomic uint64_t ctrl_repl_lag_lsn = 0;
|
|
|
|
/* ── Descriptor table ───────────────────────────────────────────────── */
|
|
const ctrl_metric_desc_t ctrl_metric_descriptors[] = {
|
|
{ "ctrl_repl_degraded_count",
|
|
&ctrl_repl_degraded_count,
|
|
"Sync replication fallbacks to local-only durability",
|
|
1 },
|
|
{ "ctrl_split_brain_detected",
|
|
&ctrl_split_brain_detected,
|
|
"Term conflicts indicating split-brain risk",
|
|
1 },
|
|
{ "ctrl_async_events_lost",
|
|
&ctrl_async_events_lost,
|
|
"Async WAL entries lost during failover",
|
|
1 },
|
|
{ "ctrl_transient_jobs_stuck",
|
|
&ctrl_transient_jobs_stuck,
|
|
"TRANSIENT jobs auto-terminated after timeout",
|
|
1 },
|
|
{ "ctrl_wal_bytes",
|
|
&ctrl_wal_bytes,
|
|
"Current WAL size in bytes",
|
|
0 },
|
|
{ "ctrl_checkpoint_duration_ms",
|
|
&ctrl_checkpoint_duration_ms,
|
|
"Duration of last checkpoint in milliseconds",
|
|
0 },
|
|
{ "ctrl_repl_lag_lsn",
|
|
&ctrl_repl_lag_lsn,
|
|
"LSN gap between primary and slowest standby",
|
|
0 },
|
|
{ NULL, NULL, NULL, 0 } /* sentinel */
|
|
};
|
|
|
|
/* ── ctrl_metrics_init ──────────────────────────────────────────────── */
|
|
void ctrl_metrics_init(void)
|
|
{
|
|
atomic_store_explicit(&ctrl_repl_degraded_count, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_split_brain_detected, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_async_events_lost, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_transient_jobs_stuck, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_wal_bytes, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_checkpoint_duration_ms, 0, memory_order_relaxed);
|
|
atomic_store_explicit(&ctrl_repl_lag_lsn, 0, memory_order_relaxed);
|
|
}
|