diff options
author | yngveaasheim <yngve@yahooinc.com> | 2023-07-31 14:10:38 +0200 |
---|---|---|
committer | yngveaasheim <yngve@yahooinc.com> | 2023-07-31 14:10:38 +0200 |
commit | 5ab417895e48ba4b45d624e899a80e866d25a442 (patch) | |
tree | c3a8ab132a80650419ff4d0ff004ffd2654f929b /metrics | |
parent | d6a8c56d1816d8e6c793c9dc3e2e527fad09e27c (diff) |
Document and add some more infrastructure metrics
Diffstat (limited to 'metrics')
3 files changed, 24 insertions, 1 deletions
diff --git a/metrics/src/main/java/ai/vespa/metrics/ConfigServerMetrics.java b/metrics/src/main/java/ai/vespa/metrics/ConfigServerMetrics.java index a277f6b4dda..ca028547171 100644 --- a/metrics/src/main/java/ai/vespa/metrics/ConfigServerMetrics.java +++ b/metrics/src/main/java/ai/vespa/metrics/ConfigServerMetrics.java @@ -112,6 +112,7 @@ public enum ConfigServerMetrics implements VespaMetrics { HOSTED_VESPA_DOCKER_ALLOCATED_CAPACITY_DISK("hostedVespa.docker.allocatedCapacityDisk", Unit.GIGABYTE, "Total amount of allocated disk space on tenant hosts managed by hosted Vespa in a zone"), HOSTED_VESPA_PENDING_REDEPLOYMENTS("hostedVespa.pendingRedeployments", Unit.TASK, "The number of hosted Vespa re-deployments pending"), HOSTED_VESPA_DOCKER_SKEW("hostedVespa.docker.skew", Unit.FRACTION, "A number in the range 0..1 indicating how well allocated resources are balanced with availability on hosts"), + HOSTED_VESPA_ACTIVE_HOSTS("hostedVespa.activeHosts", Unit.HOST, "The number of managed hosts that are in state \"active\""), HOSTED_VESPA_BREAKFIXED_HOSTS("hostedVespa.breakfixedHosts", Unit.HOST, "The number of managed hosts that are in state \"breakfixed\""), HOSTED_VESPA_DEPROVISIONED_HOSTS("hostedVespa.deprovisionedHosts", Unit.HOST, "The number of managed hosts that are in state \"deprovisioned\""), @@ -123,6 +124,17 @@ public enum ConfigServerMetrics implements VespaMetrics { HOSTED_VESPA_READY_HOSTS("hostedVespa.readyHosts", Unit.HOST, "The number of managed hosts that are in state \"ready\""), HOSTED_VESPA_RESERVED_HOSTS("hostedVespa.reservedHosts", Unit.HOST, "The number of managed hosts that are in state \"reserved\""), + HOSTED_VESPA_ACTIVE_NODES("hostedVespa.activeNodes", Unit.HOST, "The number of managed nodes that are in state \"active\""), + HOSTED_VESPA_BREAKFIXED_NODES("hostedVespa.breakfixedNodes", Unit.HOST, "The number of managed nodes that are in state \"breakfixed\""), + HOSTED_VESPA_DEPROVISIONED_NODES("hostedVespa.deprovisionedNodes", Unit.HOST, "The number of managed nodes that are in state \"deprovisioned\""), + HOSTED_VESPA_DIRTY_NODES("hostedVespa.dirtyNodes", Unit.HOST, "The number of managed nodes that are in state \"dirty\""), + HOSTED_VESPA_FAILED_NODES("hostedVespa.failedNodes", Unit.HOST, "The number of managed nodes that are in state \"failed\""), + HOSTED_VESPA_INACTIVE_NODES("hostedVespa.inactiveNodes", Unit.HOST, "The number of managed nodes that are in state \"inactive\""), + HOSTED_VESPA_PARKED_NODES("hostedVespa.parkedNodes", Unit.HOST, "The number of managed nodes that are in state \"parked\""), + HOSTED_VESPA_PROVISIONED_NODES("hostedVespa.provisionedNodes", Unit.HOST, "The number of managed nodes that are in state \"provisioned\""), + HOSTED_VESPA_READY_NODES("hostedVespa.readyNodes", Unit.HOST, "The number of managed nodes that are in state \"ready\""), + HOSTED_VESPA_RESERVED_NODES("hostedVespa.reservedNodes", Unit.HOST, "The number of managed nodes that are in state \"reserved\""), + OVERCOMMITTED_HOSTS("overcommittedHosts", Unit.HOST, "The number of hosts with over-committed resources"), SPARE_HOST_CAPACITY("spareHostCapacity", Unit.HOST, "The number of spare hosts"), diff --git a/metrics/src/main/java/ai/vespa/metrics/ControllerMetrics.java b/metrics/src/main/java/ai/vespa/metrics/ControllerMetrics.java index aa2ebd62284..01146adb8f0 100644 --- a/metrics/src/main/java/ai/vespa/metrics/ControllerMetrics.java +++ b/metrics/src/main/java/ai/vespa/metrics/ControllerMetrics.java @@ -46,7 +46,16 @@ public enum ControllerMetrics implements VespaMetrics { OPERATION_FLAGS("operation.flags", Unit.REQUEST, "Controller: Requests for /flags API"), OPERATION_OS("operation.os", Unit.REQUEST, "Controller: Requests for /os API"), OPERATION_ROUTING("operation.routing", Unit.REQUEST, "Controller: Requests for /routing API"), - OPERATION_ZONE("operation.zone", Unit.REQUEST, "Controller: Requests for /zone API"); + OPERATION_ZONE("operation.zone", Unit.REQUEST, "Controller: Requests for /zone API"), + + // Metering metrics - not used - TODO: remove from controller code. + METERING_AGE_SECONDS("metering.age.seconds", Unit.SECOND, "Controller: Metering age seconds"), + METERING_COST_HOURLY("metering.cost.hourly", Unit.DOLLAR_PER_HOUR, "Controller: Metering cost hourly"), + METERING_DISK_GB("metering.diskGB", Unit.GIGABYTE, "Controller: Metering disk GB"), + METERING_MEMORY_GB("metering.memoryGB", Unit.GIGABYTE, "Controller: Metering memory GB"), + METERING_VCPU("metering.vcpu", Unit.VCPU, "Controller: Metering VCPU"), + METERING_LAST_REPORTED("metering_last_reported", Unit.SECONDS_SINCE_EPOCH, "Controller: Metering last reported"), + METERING_TOTAL_REPORTED("metering_total_reported", Unit.ITEM, "Controller: Metering total reported (sum of resources)"); private final String name; diff --git a/metrics/src/main/java/ai/vespa/metrics/Unit.java b/metrics/src/main/java/ai/vespa/metrics/Unit.java index 3a438efde06..48b4e72891f 100644 --- a/metrics/src/main/java/ai/vespa/metrics/Unit.java +++ b/metrics/src/main/java/ai/vespa/metrics/Unit.java @@ -40,6 +40,7 @@ public enum Unit { ROTATION(BaseUnit.ROTATION, "Routing rotation"), SCORE(BaseUnit.SCORE, "Relevance score for a document"), SECOND(BaseUnit.SECOND, "Time span of 1 second"), + SECONDS_SINCE_EPOCH(BaseUnit.SECONDS_SINCE_EPOCH,"Seconds since Unix Epoch"), SESSION(BaseUnit.SESSION, "A set of operations taking place during one connection or as part of a higher level operation"), TASK(BaseUnit.TASK, "Piece of work executed by a server, e.g. to perform back-ground data maintenance"), TENANT(BaseUnit.TENANT, "Tenant that owns zero or more applications in a managed Vespa system"), @@ -115,6 +116,7 @@ public enum Unit { ROTATION("routing rotation"), SCORE("score"), SECOND("second", "s"), + SECONDS_SINCE_EPOCH("seconds since epoch"), SESSION("session"), TASK("task"), TENANT("tenant"), |