Skip to content

SYSTEM Cited by 1 source

Pinterest Goku (in-house TSDB)

Goku is Pinterest's in-house time-series datastore, built to replace HBase + OpenTSDB for time-series workloads as part of Pinterest's workload-specific migration off HBase.

Definition

Goku is described in a separate Pinterest Engineering post ("Goku: Building a Scalable and High-Performant Time Series Database System at Pinterest") that is linked from the HBase deprecation retrospective but not ingested on the wiki. The system provides:

  • A purpose-built TSDB scaled for Pinterest's monitoring data volume.
  • Performance characteristics tuned for time-series access patterns (ingestion throughput, range queries, downsampling) rather than generic wide-column workloads.

Role in HBase deprecation

Goku is the time-series-axis target of patterns/workload-specific-datastore-migration at Pinterest:

"It was not able to keep up with the ever increasing time series data volume, which led to significant challenges in scalability, performance, and maintenance load. [...] time series data [migrated] to Goku, an in-house time-series datastore." (Source: sources/2024-05-14-pinterest-hbase-deprecation-at-pinterest)

Moving time-series off HBase:

  • Shrinks the HBase footprint (one of many axes that reduced before the formal deprecation).
  • Gives time-series workloads a substrate engineered for their shape.
  • Eliminates the OpenTSDB-on-HBase operational cost (two layers of systems to maintain for one workload).

Seen in

Last updated · 550 distilled / 1,221 read