refactor(cu): renamed total evaluation gauge and finer labels for fin…

…er time-series #917
permaweb · Aug 6, 2024 · e4c79f7 · e4c79f7
1 parent b158429
commit e4c79f7
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 12 deletions.
diff --git a/servers/cu/src/domain/api/dryRun.js b/servers/cu/src/domain/api/dryRun.js
@@ -81,12 +81,7 @@ export function dryRunWith (env) {
            * Otherwise, we are evaluating up to the latest
            */
           exact: !!messageTxId,
-          needsMemory: true,
-          /**
-           * Add an indicator that this eval stream is part of a dry run.
-           * This is used for metric purposes.
-           */
-          dryRun: true
+          needsMemory: true
         })
       )
       /**

diff --git a/servers/cu/src/domain/api/readState.js b/servers/cu/src/domain/api/readState.js
@@ -35,7 +35,7 @@ export function readStateWith (env) {
   const loadModule = loadModuleWith(env)
   const evaluate = evaluateWith(env)
 
-  return ({ processId, messageId, to, ordinate, cron, exact, needsMemory, dryRun = false }) => {
+  return ({ processId, messageId, to, ordinate, cron, exact, needsMemory }) => {
     messageId = messageId || [to, ordinate, cron].filter(isNotNil).join(':') || 'latest'
 
     const stats = {
@@ -79,7 +79,7 @@ export function readStateWith (env) {
          * there is only one instance of the work used to resolve each Async,
          * every time, thus preventing duplication of work
          */
-        pending = of({ id: processId, messageId, to, ordinate, cron, stats, needsMemory, dryRun })
+        pending = of({ id: processId, messageId, to, ordinate, cron, stats, needsMemory })
           .chain(loadProcessMeta)
           .chain(loadProcess)
           .chain(loadModule)

diff --git a/servers/cu/src/domain/index.js b/servers/cu/src/domain/index.js
@@ -239,9 +239,9 @@ export const createApis = async (ctx) => {
   }
 
   const evaluationCounter = MetricsClient.counterWith({})({
-    name: 'total_evaluations',
+    name: 'ao_process_total_evaluations',
     description: 'The total number of evaluations on a CU',
-    labelNames: ['processId', 'cron', 'dryRun', 'error']
+    labelNames: ['stream_type', 'message_type', 'process_error']
   })
 
   /**

diff --git a/servers/cu/src/domain/lib/evaluate.js b/servers/cu/src/domain/lib/evaluate.js
@@ -241,9 +241,21 @@ export function evaluateWith (env) {
                           }
 
                           /**
-                           * Increments gauges for total evaluations
+                           * Increments gauges for total evaluations for both:
+                           *
+                           * total evaluations (no labels)
+                           * specific kinds of evaluations (type of eval stream, type of message, whether an error occurred or not)
                            */
-                          evaluationCounter.inc(1, { cron: Boolean(cron), dryRun: Boolean(ctx.dryRun) }, { processId: ctx.id, error: Boolean(output.Error) })
+                          evaluationCounter.inc(1)
+                          evaluationCounter.inc(
+                            1,
+                            {
+                              stream_type: ctx.dryRun ? 'dry-run' : 'primary',
+                              message_type: ctx.dryRun ? 'dry-run' : cron ? 'cron' : isAssignment ? 'assignment' : 'scheduled',
+                              process_error: Boolean(output.Error)
+                            },
+                            { processId: ctx.id }
+                          )
                           /**
                            * TODO: Gas can grow to a huge number. We need to make sure this doesn't crash when that happens
                            */