jam
10/12/2022, 3:50 PMpat_nadolny
10/12/2022, 4:00 PMjam
10/12/2022, 4:08 PMjam
10/12/2022, 4:08 PMedgar_ramirez_mondragon
10/12/2022, 4:57 PMIs there a way to load from tap -> multiple targets with a single meltano command?@jam There isn’t at the moment. As far as I can tell, the main impediment for that is the coupling of tap & target created by the state. Even if you could create a fan-out that delivers singer messages to multiple targets, if one of the targets fails, and the other continues, they’ll end up with different states. The user would be forced to have separate invocations of the tap in the next run to use the different starting states.
jam
10/12/2022, 5:07 PMjam
10/12/2022, 5:55 PMThis capability is currently in development by the Singer Working Group to align on a consistent approach for supporting batch sync messages in the Spec or as part of its officially supported patterns. The capability, which was inspired by Wise’s Fast Sync feature, has the goal of bypassing some of the performance limitations inherent to the Singer Spec by using the optimized import/export features of modern data warehouses (Snowflake, Redshift, etc.). For example if a tap and target both had batch support the tap could theoretically know to directly write records to csv files somewhere (i.e. S3) in an ideal format for the target to import, ultimately skipping most of the piping step. Taking this approach allows connectors to improve performance by both using the source and/or target’s most optimized technique for importing and exporting data while also avoiding the additional costs of serializing, deserializing, and piping each record. Join the issue conversation here!https://hub.meltano.com/singer/docs/#batch
pat_nadolny
10/12/2022, 6:27 PMjam
10/12/2022, 6:28 PMpat_nadolny
10/12/2022, 6:41 PMjam
10/12/2022, 6:42 PMedgar_ramirez_mondragon
10/12/2022, 6:44 PMaaronsteers
10/12/2022, 6:47 PMretain_s3_files
for backup purposes. The nice thing about this approach is that the target still gets full fidelity of the upstream tap's catalog metadata, where otherwise some data types would morph if doing this as a two-step process. (Scanning metadata like data types on the S3 files might have less fidelity than what you'd get from the tap itself.)pat_nadolny
10/12/2022, 6:47 PMaaronsteers
10/12/2022, 6:48 PMaaronsteers
10/12/2022, 6:49 PMjam
10/12/2022, 6:52 PMaaronsteers
10/12/2022, 7:44 PMaaronsteers
10/12/2022, 7:45 PMjam
10/13/2022, 4:46 PMaaronsteers
10/13/2022, 11:18 PM