Support sub-second intervals in hierarchical caggs

Internally we use date_part("epoch", interval) and integer division
to determine whether the top cagg's interval is a multiple of
the parent's.
This would lead to loss of precision and incorrect results in the case
of intervals with sub-second components.
Fixed by multiplying by appropriate factors when possible to maintain
precision.

Fixes timescale#5277

CHANGELOG.md

@@ -18,6 +18,7 @@ accidentally triggering the load of a previous DB version.**
 * #5214 Fix use of prepared statement in async module
 * #5290 Compression can't be enabled on continuous aggregates when segmentby/orderby columns need quotation
 * #5239 Fix next_start calculation for fixed schedules
+* #5304 Support sub-second intervals in hierarchical caggs
 
 ## 2.9.3 (2023-02-03)
 

tsl/src/continuous_aggs/create.c

@@ -1147,9 +1147,19 @@ get_bucket_width(CAggTimebucketInfo bucket_info)
 				bucket_info.interval->day = bucket_info.interval->month * DAYS_PER_MONTH;
 				bucket_info.interval->month = 0;
 			}
+			int64 MSECS_PER_SEC = 1000;
+			Datum float_usecs_per_sec = DirectFunctionCall1(i8tod, Int64GetDatum(USECS_PER_SEC));
+			Datum float_msecs_per_sec = DirectFunctionCall1(i8tod, Int64GetDatum(MSECS_PER_SEC));
 			Datum epoch = DirectFunctionCall2(interval_part,
 											  PointerGetDatum(cstring_to_text("epoch")),
 											  IntervalPGetDatum(bucket_info.interval));
+			/* Preserve precision if possible for units less than seconds by converting to usecs or
+			 * msecs */
+			if (PG_INT64_MAX / USECS_PER_SEC + PG_INT64_MAX % USECS_PER_SEC > DatumGetFloat8(epoch))
+				epoch = DirectFunctionCall2(float8mul, epoch, float_usecs_per_sec);
+			else if (PG_INT64_MAX / MSECS_PER_SEC + PG_INT64_MAX % MSECS_PER_SEC >
+					 DatumGetFloat8(epoch))
+				epoch = DirectFunctionCall2(float8mul, epoch, float_msecs_per_sec);
 			/* Cast float8 to int8. */
 			width = DatumGetInt64(DirectFunctionCall1(dtoi8, epoch));
 			break;