I tried to run in-memory mode and pushed thousands of metrics to the storage.
The datapoints span more then 1d - about 10 metrics 1 each minute.
partition-size is 1h which is the default, retention is 14d which is the default.
I've noticed that once the 3rd partitionList is generated the following method will be invoked flushPartitions and the following code will delete the older partition:
if s.inMemoryMode() {
if err := s.partitionList.remove(part); err != nil {
return fmt.Errorf("failed to remove partition: %w", err)
}
continue
}
This will cause Select to fail on metrics that were part of the first partition that was removed.
I don't understand what does flush mean for in-memory. I guess it should be no-op and the only method that should delete partitions is: removeExpiredPartitions
I understand for disk storage we would like to flush the partition to disk for persistence but the current code will not hold more then 2 partitions in-memory at all times.
I tried to run in-memory mode and pushed thousands of metrics to the storage.
The datapoints span more then
1d- about 10 metrics 1 each minute.partition-size is
1hwhich is the default, retention is 14d which is the default.I've noticed that once the 3rd partitionList is generated the following method will be invoked
flushPartitionsand the following code will delete the older partition:This will cause Select to fail on metrics that were part of the first partition that was removed.
I don't understand what does flush mean for in-memory. I guess it should be no-op and the only method that should delete partitions is:
removeExpiredPartitionsI understand for disk storage we would like to flush the partition to disk for persistence but the current code will not hold more then 2 partitions in-memory at all times.