Drop synapse_storage_transaction_time_bucket

This particular metric has a much too high cardinality due to the fact
that the desc label can have (at present) 248 values. This results in
over 3k series per Synapse. If you have a Prometheus instance that
monitors multiple Synpase instances it results in a huge amount of
additional series to ingest.

The metric in question is also not used in the Synapse dashboard and
the core team has indicated they're happy to drop this metric entirely.

Fixes #11081

Signed-off-by: Daniele Sluijters <daenney@users.noreply.github.com>

changelog.d/11124.removal

@@ -0,0 +1 @@
+Remove the `synapse_storage_transaction_time_bucket` metric due to the high cardinality of the metric putting undue strain on Prometheus deployments. This metric is not used in Synapse's included Grafana dashboards.

synapse/storage/database.py

@@ -64,7 +64,6 @@
 sql_scheduling_timer = Histogram("synapse_storage_schedule_time", "sec")
 
 sql_query_timer = Histogram("synapse_storage_query_time", "sec", ["verb"])
-sql_txn_timer = Histogram("synapse_storage_transaction_time", "sec", ["desc"])
 
 
 # Unique indexes which have been added in background updates. Maps from table name
@@ -639,7 +638,6 @@ def new_transaction(
 
             self._current_txn_total_time += duration
             self._txn_perf_counters.update(desc, duration)
-            sql_txn_timer.labels(desc).observe(duration)
 
     async def runInteraction(
         self,