Fix tests and exposition

Two things:
 * Changing to a GaugeHistogram, and adding help output, mildly changed the
   output from the metrics, so we need to update the tests
 * Our hacked-up `exposition.py` was stripping out some samples it shouldn't
   have been. Put them back in, to more closely match the upstream
   `exposition.py`.

synapse/metrics/_exposition.py

@@ -124,16 +124,33 @@ def generate_latest(registry, emit_help=False):
                 )
             )
         output.append("# TYPE {0} {1}\n".format(mname, mtype))
-        for sample in metric.samples:
-            # Get rid of the OpenMetrics specific samples
+
+        om_samples = {}
+        for s in metric.samples:
             for suffix in ["_created", "_gsum", "_gcount"]:
-                if sample.name.endswith(suffix):
+                if s.name == metric.name + suffix:
+                    # OpenMetrics specific sample, put in a gauge at the end.
+                    # (these come from gaugehistograms which don't get renamed,
+                    # so no need to faff with mnewname)
+                    om_samples.setdefault(suffix, []).append(sample_line(s, s.name))
                     break
             else:
-                newname = sample.name.replace(mnewname, mname)
+                newname = s.name.replace(mnewname, mname)
                 if ":" in newname and newname.endswith("_total"):
                     newname = newname[: -len("_total")]
-                output.append(sample_line(sample, newname))
+                output.append(sample_line(s, newname))
+
+        for suffix, lines in sorted(om_samples.items()):
+            if emit_help:
+                output.append(
+                    "# HELP {0}{1} {2}\n".format(
+                        metric.name,
+                        suffix,
+                        metric.documentation.replace("\\", r"\\").replace("\n", r"\n"),
+                    )
+                )
+            output.append("# TYPE {0}{1} gauge\n".format(metric.name, suffix))
+            output.extend(lines)
 
         # Get rid of the weird colon things while we're at it
         if mtype == "counter":
@@ -152,16 +169,16 @@ def generate_latest(registry, emit_help=False):
                 )
             )
         output.append("# TYPE {0} {1}\n".format(mnewname, mtype))
-        for sample in metric.samples:
-            # Get rid of the OpenMetrics specific samples
+
+        for s in metric.samples:
+            # Get rid of the OpenMetrics specific samples (we should already have
+            # dealt with them above anyway.)
             for suffix in ["_created", "_gsum", "_gcount"]:
-                if sample.name.endswith(suffix):
+                if s.name == metric.name + suffix:
                     break
             else:
                 output.append(
-                    sample_line(
-                        sample, sample.name.replace(":total", "").replace(":", "_")
-                    )
+                    sample_line(s, s.name.replace(":total", "").replace(":", "_"))
                 )
 
     return "".join(output).encode("utf-8")

tests/storage/test_event_metrics.py

@@ -52,14 +52,14 @@ def test_exposed_to_prometheus(self):
         self.reactor.advance(60 * 60 * 1000)
         self.pump(1)
 
-        items = set(
+        items = list(
             filter(
                 lambda x: b"synapse_forward_extremities_" in x,
-                generate_latest(REGISTRY).split(b"\n"),
+                generate_latest(REGISTRY, emit_help=False).split(b"\n"),
             )
         )
 
-        expected = {
+        expected = [
             b'synapse_forward_extremities_bucket{le="1.0"} 0.0',
             b'synapse_forward_extremities_bucket{le="2.0"} 2.0',
             b'synapse_forward_extremities_bucket{le="3.0"} 2.0',
@@ -72,9 +72,12 @@ def test_exposed_to_prometheus(self):
             b'synapse_forward_extremities_bucket{le="100.0"} 3.0',
             b'synapse_forward_extremities_bucket{le="200.0"} 3.0',
             b'synapse_forward_extremities_bucket{le="500.0"} 3.0',
-            b'synapse_forward_extremities_bucket{le="+Inf"} 3.0',
-            b"synapse_forward_extremities_count 3.0",
-            b"synapse_forward_extremities_sum 10.0",
-        }
-
+            # per https://docs.google.com/document/d/1KwV0mAXwwbvvifBvDKH_LU1YjyXE_wxCkHNoCGq1GX0/edit#heading=h.wghdjzzh72j9,
+            # "inf" is valid: "this includes variants such as inf"
+            b'synapse_forward_extremities_bucket{le="inf"} 3.0',
+            b"# TYPE synapse_forward_extremities_gcount gauge",
+            b"synapse_forward_extremities_gcount 3.0",
+            b"# TYPE synapse_forward_extremities_gsum gauge",
+            b"synapse_forward_extremities_gsum 10.0",
+        ]
         self.assertEqual(items, expected)