Module metrics 

Prometheus metrics for the process-wide buffer pool and columnar chunk work.

Installed once by compute init via register. All metrics are computed gauges that peek at the pool’s stats at scrape time, reporting zero until something initializes the pool — a scrape must observe, not mmap the pool’s address space into every process that happens to be monitored.

Functions

backend_gauges 🔒

Registers mz_column_pool_backend, one gauge per backend kind that reads 1 for the kind the pool runs on and 0 otherwise, including before the pool exists.

gauge 🔒

Registers one computed gauge over a mz_ore::pool::PoolStats field. Peeks at the process-wide pool at scrape time. Reports zero until something initializes the pool, or if its reservation failed.

read_latency_gauges 🔒

Registers the file store’s read-latency histogram as one computed gauge per bucket, since the registry has no computed histogram. Each gauge carries an le label, the bucket’s upper bound in seconds, and counts the reads under that bound, so the series aggregate like the buckets of a Prometheus histogram. Bucket i below the last holds reads under 32 µs << i, and the last bucket is +Inf.

register

Install the buffer-pool metrics and the columnar chunk work counters into registry. Idempotent. Repeated calls after the first one are no-ops.