SERVER-66727 Use full-precision timestamp for time-series bucketing decision

2 files changed, 16 insertions, 9 deletions

diff --git a/src/mongo/db/timeseries/bucket_catalog.cpp b/src/mongo/db/timeseries/bucket_catalog.cpp
index 461a90bb213..7093a63d0eb 100644
--- a/src/mongo/db/timeseries/bucket_catalog.cpp
+++ b/src/mongo/db/timeseries/bucket_catalog.cpp
@@ -166,7 +166,7 @@ std::pair<OID, Date_t> generateBucketId(const Date_t& time, const TimeseriesOpti
     // together into predictable chunks for sharding. This way we know from a measurement timestamp
     // what the bucket timestamp will be, so we can route measurements to the right shard chunk.
     auto roundedTime = timeseries::roundTimestampToGranularity(time, options.getGranularity());
-    uint64_t const roundedSeconds = durationCount<Seconds>(roundedTime.toDurationSinceEpoch());
+    int64_t const roundedSeconds = durationCount<Seconds>(roundedTime.toDurationSinceEpoch());
     bucketId.setTimestamp(roundedSeconds);
 
     // Now, if we stopped here we could end up with bucket OID collisions. Consider the case where
@@ -265,7 +265,7 @@ StatusWith<std::shared_ptr<BucketCatalog::WriteBatch>> BucketCatalog::insert(
             stats->numBucketsClosedDueToSize.fetchAndAddRelaxed(1);
             return true;
         }
-        auto bucketTime = (*bucket).getTime();
+        auto bucketTime = (*bucket)->getTime();
         if (time - bucketTime >= Seconds(*options.getBucketMaxSpanSeconds())) {
             stats->numBucketsClosedDueToTimeForward.fetchAndAddRelaxed(1);
             return true;
@@ -671,6 +671,8 @@ BucketCatalog::Bucket* BucketCatalog::_allocateBucket(const BucketKey& key,
         stats->numBucketsOpenedDueToMetadata.fetchAndAddRelaxed(1);
     }
 
+    bucket->_minTime = roundedTime;
+
     // Make sure we set the control.min time field to match the rounded _id timestamp.
     auto controlDoc = buildControlMinTimestampDoc(options.getTimeField(), roundedTime);
     bucket->_minmax.update(
@@ -851,6 +853,10 @@ bool BucketCatalog::Bucket::allCommitted() const {
     return _batches.empty() && !_preparedBatch;
 }
 
+Date_t BucketCatalog::Bucket::getTime() const {
+    return _minTime;
+}
+
 std::shared_ptr<BucketCatalog::WriteBatch> BucketCatalog::Bucket::_activeBatch(
     OperationId opId, const std::shared_ptr<ExecutionStats>& stats) {
     auto it = _batches.find(opId);
@@ -1126,10 +1132,6 @@ void BucketCatalog::BucketAccess::rollover(const std::function<bool(BucketAccess
     }
 }
 
-Date_t BucketCatalog::BucketAccess::getTime() const {
-    return _bucket->id().asDateT();
-}
-
 BucketCatalog::WriteBatch::WriteBatch(const OID& bucketId,
                                       OperationId opId,
                                       const std::shared_ptr<ExecutionStats>& stats)
diff --git a/src/mongo/db/timeseries/bucket_catalog.h b/src/mongo/db/timeseries/bucket_catalog.h
index f00475a0c71..59266a6d765 100644
--- a/src/mongo/db/timeseries/bucket_catalog.h
+++ b/src/mongo/db/timeseries/bucket_catalog.h
@@ -349,6 +349,11 @@ public:
          */
         bool allCommitted() const;
 
+        /**
+         * Retrieve the time associated with the bucket
+         */
+        Date_t getTime() const;
+
     private:
         /**
          * Determines the effect of adding 'doc' to this bucket. If adding 'doc' causes this bucket
@@ -393,6 +398,9 @@ public:
         // Top-level field names of the measurements that have been inserted into the bucket.
         StringSet _fieldNames;
 
+        // Minimum timestamp over contained measurements
+        Date_t _minTime;
+
         // The minimum and maximum values for each field in the bucket.
         timeseries::MinMax _minmax;
 
@@ -575,9 +583,6 @@ private:
          */
         void rollover(const std::function<bool(BucketAccess*)>& isBucketFull);
 
-        // Retrieve the time associated with the bucket (id)
-        Date_t getTime() const;
-
     private:
         /**
          * Returns the state of the bucket, or boost::none if there is no state for the bucket.