add user experience related metrics

src/ClusterManager/job_manager.py

@@ -41,6 +41,90 @@
             float("inf")),
         labelnames=("fn_name",))
 
+job_state_change_histogram = Histogram("job_state_change_latency_seconds",
+        "latency for job to change state(seconds)",
+        buckets=(1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0,
+            float("inf")),
+        labelnames=("current_state",))
+
+class JobTimeRecord(object):
+    def __init__(self):
+        self.create_time = None
+        self.approve_time = None
+        self.submit_time = None
+
+
+class LRUDefatulDict(object):
+    class Node(object):
+        def __init__(self, key, val, next=None, prev=None):
+            self.key = key
+            self.val = val
+            self.next = next
+            self.prev = prev
+
+    def __init__(self, cap, factory):
+        assert cap > 0
+        self.m = {}
+        self.head = self.tail = None
+        self.cap = cap
+        self.factory = factory
+
+    def __getitem__(self, key):
+        return self._get(key)
+
+    def _get(self, key):
+        if self.m.get(key) is None:
+            self._put(key, self.factory())
+
+        return self.m[key].val
+
+    def _put(self, key, value):
+        if self.m.get(key) is not None:
+            node = self.m[key]
+            if node == self.head:
+                node.val = value
+            else:
+                node.prev.next = node.next
+                if node.next is not None:
+                    node.next.prev = node.prev
+                node.prev = None
+                node.next = self.head
+                self.head.prev = node
+                self.head = node
+        else:
+            self.head = LRUDefatulDict.Node(key, value, self.head, None)
+            if self.tail is None:
+                self.tail = self.head
+            else:
+                self.head.next.prev = self.head
+            self.m[key] = self.head
+            if len(self.m) > self.cap:
+                self.m.pop(self.tail.key)
+                self.tail = self.tail.prev
+                self.tail.next = None
+
+# pure memory data structure
+job_time_recorder = LRUDefatulDict(500, lambda : JobTimeRecord())
+
+# If previous state has no record, which means the job_manager get restarted
+# or previous entry is expired, we ignore this entry.
+def update_job_state_latency(job_id, state, event_time=None):
+    if event_time is None:
+        event_time = datetime.datetime.utcnow()
+
+    if state == "create":
+        job_time_recorder[job_id].create_time = event_time
+    elif state == "approve":
+        job_time_recorder[job_id].approve_time = event_time
+        if job_time_recorder[job_id].create_time is not None:
+            elapsed = (event_time - job_time_recorder[job_id].create_time).seconds
+            job_state_change_histogram.labels(state).observe(elapsed)
+    elif state == "submit":
+        job_time_recorder[job_id].submit_time = event_time
+        if job_time_recorder[job_id].approve_time is not None:
+            elapsed = (event_time - job_time_recorder[job_id].approve_time).seconds
+            job_state_change_histogram.labels(state).observe(elapsed)
+
 def record(fn):
     @functools.wraps(fn)
     def wrapped(*args, **kwargs):
@@ -98,6 +182,9 @@ def ApproveJob(job, dataHandlerOri=None):
     try:
         job_id = job["jobId"]
         vcName = job["vcName"]
+
+        update_job_state_latency(job_id, "create", event_time=job["jobTime"])
+
         jobParams = json.loads(base64.b64decode(job["jobParams"]))
         job_total_gpus = GetJobTotalGpu(jobParams)
 
@@ -111,6 +198,7 @@ def ApproveJob(job, dataHandlerOri=None):
             detail = [{"message": "waiting for available preemptible resource."}]
             dataHandler.UpdateJobTextField(job["jobId"], "jobStatusDetail", base64.b64encode(json.dumps(detail)))
             dataHandler.UpdateJobTextField(job_id, "jobStatus", "queued")
+            update_job_state_latency(job_id, "approve")
             if dataHandlerOri is None:
                 dataHandler.Close()
             return True
@@ -151,6 +239,7 @@ def ApproveJob(job, dataHandlerOri=None):
         detail = [{"message": "waiting for available resource."}]
         dataHandler.UpdateJobTextField(job["jobId"], "jobStatusDetail", base64.b64encode(json.dumps(detail)))
         dataHandler.UpdateJobTextField(job_id, "jobStatus", "queued")
+        update_job_state_latency(job_id, "approve")
         if dataHandlerOri is None:
             dataHandler.Close()
         return True
@@ -443,6 +532,7 @@ def TakeJobActions(launcher, jobs):
         try:
             if sji["job"]["jobStatus"] == "queued" and (sji["allowed"] is True):
                 launcher.submit_job(sji["job"])
+                update_job_state_latency(sji["jobId"], "submit")
                 logging.info("TakeJobActions : submitting job : %s : %s" % (sji["jobId"], sji["sortKey"]))
             elif sji["preemptionAllowed"] and (sji["job"]["jobStatus"] == "scheduling" or sji["job"]["jobStatus"] == "running") and (sji["allowed"] is False):
                 launcher.kill_job(sji["job"]["jobId"], "queued")