Rework the way metrics are handled

Start writing a proper Prometheus client, hopefully this code can be extracted
at some point.

1 files changed, 79 insertions, 4 deletions

diff --git a/guix-build-coordinator/agent-messaging/http.scm b/guix-build-coordinator/agent-messaging/http.scm
index 6c5bf51..4b569f1 100644
--- a/guix-build-coordinator/agent-messaging/http.scm
+++ b/guix-build-coordinator/agent-messaging/http.scm
@@ -40,6 +40,7 @@
   #:use-module (guix serialization)
   #:use-module (guix build utils)
   #:use-module (guix-build-coordinator utils)
+  #:use-module (guix-build-coordinator metrics)
   #:use-module (guix-build-coordinator datastore)
   #:use-module (guix-build-coordinator metrics)
   #:use-module (guix-build-coordinator coordinator)
@@ -86,6 +87,10 @@ if there was no request body."
     (make-worker-thread-channel (const '())
                                 #:parallelism 4))
 
+  (define update-base-datastore-metrics!
+    (base-datastore-metrics-updater datastore
+                                    coordinator-metrics-registry))
+
   (trigger-build-allocation)
   (call-with-error-handling
    (lambda ()
@@ -105,7 +110,8 @@ if there was no request body."
                            datastore
                            trigger-build-allocation
                            hook-channel
-                           chunked-request-channel)))
+                           chunked-request-channel
+                           update-base-datastore-metrics!)))
       #:host host
       #:port port))
    #:on-error 'backtrace
@@ -146,6 +152,68 @@ port. Also, the port used can be changed by passing the --port option.\n"
   (list (build-response #:code 204)
         ""))
 
+(define (base-datastore-metrics-updater datastore registry)
+  (define namespace
+    "guixbuildcoordinator")
+
+  (let ((builds-total
+         (make-gauge-metric registry
+                            (string-append namespace
+                                           "_builds_total")))
+        (allocated-builds-total
+         (make-gauge-metric registry
+                            (string-append namespace
+                                           "_allocated_builds_total")
+                            #:labels '(agent_id)))
+        (build-results-total
+         (make-gauge-metric registry
+                            (string-append namespace
+                                           "_build_results_total")
+                            #:labels '(agent_id result)))
+        (setup-failures-total
+         (make-gauge-metric registry
+                            (string-append namespace
+                                           "_setup_failures_total")
+                            #:labels '(agent_id reason)))
+        (build-allocation-plan-total
+         (make-gauge-metric registry
+                            (string-append namespace
+                                           "_build_allocation_plan_total")
+                            #:labels '(agent_id))))
+    (lambda ()
+      (metric-set builds-total
+                  (datastore-count-builds datastore))
+      (for-each (match-lambda
+                  ((agent-id . count)
+                   (metric-set allocated-builds-total
+                               count
+                               #:label-values
+                               `((agent_id . ,agent-id)))))
+                (datastore-count-allocated-builds datastore))
+      (for-each (match-lambda
+                  (((agent-id result) . count)
+                   (metric-set build-results-total
+                               count
+                               #:label-values
+                               `((agent_id . ,agent-id)
+                                 (result   . ,result)))))
+                (datastore-count-build-results datastore))
+      (for-each (match-lambda
+                  (((agent-id reason) . count)
+                   (metric-set setup-failures-total
+                               count
+                               #:label-values
+                               `((agent_id . ,agent-id)
+                                 (reason   . ,reason)))))
+                (datastore-count-setup-failures datastore))
+      (for-each (match-lambda
+                  ((agent-id . count)
+                   (metric-set build-allocation-plan-total
+                               count
+                               #:label-values
+                               `((agent_id . ,agent-id)))))
+                (datastore-count-build-allocation-plan-entries  datastore)))))
+
 (define (controller request
                     method-and-path-components
                     body
@@ -153,7 +221,8 @@ port. Also, the port used can be changed by passing the --port option.\n"
                     datastore
                     trigger-build-allocation
                     hook-channel
-                    chunked-request-channel)
+                    chunked-request-channel
+                    update-base-datastore-metrics!)
   (define (authenticated? uuid request)
     (let* ((authorization-base64
             (match (assq-ref (request-headers request)
@@ -299,8 +368,14 @@ port. Also, the port used can be changed by passing the --port option.\n"
               "access denied"
               #:code 403))))
       (('GET "metrics")
-       (render-text
-        (metrics datastore)))
+       (update-base-datastore-metrics!)
+       (list (build-response
+              #:code 200
+              #:headers '((content-type . (text/plain))
+                          (vary . (accept))))
+             (lambda (port)
+               (write-metrics coordinator-metrics-registry
+                              port))))
       (_
        (render-json
         "not-found"