Track DB txn times w/ two counters, not histogram

By my reckoning, this should mean that each txn `desc` produces 2 time
series, down from 17. If we don't like this, we could keep the histogram
but configure it with coarser buckets.

Closes #11081.

synapse/storage/database.py

@@ -39,7 +39,7 @@
 )
 
 import attr
-from prometheus_client import Histogram
+from prometheus_client import Histogram, Summary, Counter
 from typing_extensions import Concatenate, Literal, ParamSpec
 
 from twisted.enterprise import adbapi
@@ -76,7 +76,8 @@
 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"])
+sql_txn_count = Counter("synapse_storage_transaction_time_count", "sec", ["desc"])
+sql_txn_duration = Counter("synapse_storage_transaction_time_sum", "sec", ["desc"])
 
 
 # Unique indexes which have been added in background updates. Maps from table name
@@ -795,7 +796,8 @@ def new_transaction(
 
             self._current_txn_total_time += duration
             self._txn_perf_counters.update(desc, duration)
-            sql_txn_timer.labels(desc).observe(duration)
+            sql_txn_count.labels(desc).inc(1)
+            sql_txn_duration.labels(desc).inc(duration)
 
     async def runInteraction(
         self,