Metric names¶
Every metric instrument emits, against its Prometheus counterpart in
go/transport-metrics. Metric names
are part of the module's compatibility contract, and
RED, USE and cardinality says why they differ.
The API is on pkg.go.dev.
Operations¶
| OTel metric | Instrument | Unit | Attributes | Prometheus metric |
|---|---|---|---|---|
operation.duration |
histogram | s |
operation.name, error.type on failure |
operation_duration_seconds |
| (its count) | operations_total |
|||
(its count with error.type) |
operation_errors_total |
|||
operation.active |
up/down counter | {operation} |
operation.name |
operations_in_flight |
DB¶
Every query metric carries db.namespace, the name given to NewDB. Every pool metric carries
db.client.connection.pool.name, the same name.
| OTel metric | Instrument | Unit | Other attributes | Prometheus metric |
|---|---|---|---|---|
db.client.operation.duration |
histogram | s |
db.operation.name, db.system.name, error.type on failure |
db_query_duration_seconds |
| (its count) | db_queries_total |
|||
(its count with error.type) |
db_query_errors_total |
|||
phpboyscout.db.client.operation.active |
up/down counter | {operation} |
db.operation.name |
db_queries_in_flight |
db.client.connection.count |
observable up/down counter | {connection} |
db.client.connection.state (used, idle) |
db_pool_in_use, db_pool_idle |
(its sum over state) |
db_pool_open |
|||
db.client.connection.max |
observable up/down counter | {connection} |
db_pool_max_open |
|
phpboyscout.db.client.connection.wait_count |
observable counter | {wait} |
db_pool_wait_count_total |
|
phpboyscout.db.client.connection.wait_time |
observable counter | s |
db_pool_wait_seconds_total |
The phpboyscout. names have no semantic convention to follow. Convention's
db.client.connection.wait_time is a histogram of individual waits, which a cumulative total
cannot feed, so phpboyscout.db.client.connection.wait_time is a counter under the same word. Only db.client.operation.duration is a Stable convention. The
db.client.connection.* metrics are in Development.
Cache¶
Every cache metric carries cache.name, the name given to NewCache.
| OTel metric | Instrument | Unit | Other attributes | Prometheus metric |
|---|---|---|---|---|
cache.operations |
counter | {operation} |
cache.result (hit, miss) |
cache_hits_total, cache_misses_total |
cache.evictions |
counter | {entry} |
cache_evictions_total |
|
cache.size |
observable gauge | {entry} |
cache_size |
|
cache.capacity |
observable gauge | {entry} |
cache_capacity |
CircuitBreaker¶
Every breaker metric carries circuit_breaker.name, the name given to NewCircuitBreaker.
| OTel metric | Instrument | Unit | Prometheus metric |
|---|---|---|---|
circuit_breaker.state |
observable gauge | {state} |
circuit_breaker_state |
circuit_breaker.trips |
counter | {trip} |
circuit_breaker_trips_total |
circuit_breaker.short_circuits |
counter | {request} |
circuit_breaker_short_circuits_total |
circuit_breaker.successes |
counter | {request} |
circuit_breaker_successes_total |
circuit_breaker.failures |
counter | {request} |
circuit_breaker_failures_total |
The state reads 0 for closed, 1 for open and 2 for half-open, as it does in transport-metrics.
Queue¶
Every queue metric carries queue.name, the name given to NewQueue.
| OTel metric | Instrument | Unit | Prometheus metric |
|---|---|---|---|
queue.depth |
up/down counter | {item} |
queue_depth |
queue.enqueued |
counter | {item} |
queue_enqueued_total |
queue.dequeued |
counter | {item} |
queue_dequeued_total |
queue.wait.duration |
histogram | s |
queue_wait_seconds |
queue.processing.duration |
histogram | s |
queue_processing_seconds |
API differences¶
transport-metrics |
instrument |
Why |
|---|---|---|
m.NewOperations(opts...) |
NewOperations(meter, opts...) |
Takes an explicit meter; there is no registry to hang a method on |
ObserveResult(o, ctx, name, fn) |
ObserveResult(ctx, o, name, fn) |
Context first, the Go convention |
m.NewDB(name, opts...) |
NewDB(meter, name, opts...) |
As above |
| (none) | WithDBSystem(name) |
Semantic conventions require db.system.name |
| (none) | DB.Close() |
An OTel callback must be unregistered, or it keeps the pool alive |
m.NewCache(name, opts...) |
NewCache(meter, name, opts...) |
As above |
cache.Hit(), Miss(), Eviction(n) |
cache.Hit(ctx), Miss(ctx), Eviction(ctx, n) |
A measurement without a context carries no exemplar |
| (none) | Cache.Close() |
An OTel callback must be unregistered |
cache.Eviction(n) panics on a negative n |
cache.Eviction(ctx, n) ignores an n below 1 |
A counter cannot go down, and a metrics call should not take a service down |
m.NewCircuitBreaker(name, opts...) |
NewCircuitBreaker(meter, name, opts...) |
As above |
cb.Trip(), ShortCircuit(), Success(), Failure() |
The same, taking ctx |
A measurement without a context carries no exemplar |
| (none) | CircuitBreaker.Close() |
An OTel callback must be unregistered |
m.NewQueue(name, opts...) |
NewQueue(meter, name, opts...) |
As above |
q.Enqueue(), Dequeue(), ObserveWait(d) |
The same, taking ctx |
A measurement without a context carries no exemplar |
WithConstLabels(labels) |
WithAttributes(attrs...) |
OTel's vocabulary |
WithNamespace(ns) |
(none) | The meter's name scopes its instruments, and a prefix would break semantic-convention names |