Decouple the controller from the data path for real-time Pinot tables.
In case of RealTime Pinot tables, whenever a Pinot server finishes consuming a segment, it goes through a segment completion protocol sequence. The default approach is to upload this segment to the lead Pinot controller which in turn will persist it in the segment store (eg: NFS, S3 or HDFS). As a result, since all the real-time segments flow through the controller, it can become a bottleneck and slow down the overall ingestion rate. To overcome this limitation, we've added a new stream-level configuration which allows bypassing the controller in the segment completion protocol.
Add the following stream-level config to allow the server to upload the completed segment to the deep store directly.
When this is enabled, the Pinot servers will attempt to upload the completed segment to the segment store directly, thus by-passing the controller. Once this is finished, it will update the controller with the corresponding segment metadata.
pinot.server.segment.store.uri
is optional by default. However, this config is required so that the server knows where the deep store is. Hence, before enabling realtime.segment.serverUploadToDeepStore
on the table, make sure that the pinot.server.segment.store.uri
is configured on the servers.
Peer download policy is introduced to allow failure recovery, incase of a failure to upload the completed segment to the deep store. If the segment store is unavailable for whatever reason, the corresponding segments can still be downloaded directly from the Pinot servers. Hence, the policy is called peer download
.
Note: This is available in the latest master (not in 0.5.0 release)
This scheme only works for real-time tables using the Low Level Consumer (LLC) mode. The changes needed are as follows:
Add the following things to the Controller Config
Add the following things to the server config
Here URI of segment store should point to the desired full path in the corresponding segment store with both filesystem scheme and path (eg: file://dir
or hdfs://path
or s3://path
)
Replace the last field (i.e., scheme) of pinot.server.storage.factory.class.(scheme) with the corresponding scheme (e.g., hdfs, s3 or gcs) of the segment store URI configured above. Then put the PinotFS subclass for the scheme as the config value.
Add the following things to the real-time segments config:
In this case, the peerSegmentDownloadScheme
can be either http
or https
.
Enabling peer download may incur LLC segments failed to be uploaded to segment store in some failure cases, e.g. segment store is unavailable during segment completion. Add the following controller config to enable the upload retry by a controller periodic job asynchronously.