ProductDocumentationExamplesBlogRoadmapGitHubGet Started
Available

Stream

krishiv stream submit / push / poll — continuous window jobs in-process.

krishiv stream runs a continuous window job entirely in the CLI process. It is intended for local development, smoke tests, and streaming CLI demos — not for production. For production, use the coordinator.

Usage

# 1. Submit a job (returns a name to use in push/poll)
krishiv stream submit --job-id events   --window tumbling --window-size-ms 60000   --key-column user_id --event-time-column ts

# 2. Push batches (Parquet files)
krishiv stream push --job-id events --parquet ./batch_001.parquet
krishiv stream push --job-id events --parquet ./batch_002.parquet

# 3. Poll for results
krishiv stream poll --job-id events

submit flags

FlagDefaultEffect
--job-id <ID> (required)Unique job ID. Use the same ID for push and poll.
--window <tumbling|sliding|session>tumblingWindow type.
--window-size-ms <MS>60000Window size.
--slide-ms <MS>30000Slide (sliding windows only).
--session-gap-ms <MS>5000Inactivity gap (session windows only).
--key-column <COL>user_idKey column for the windowed aggregation.
--event-time-column <COL>tsEvent-time column for watermarks.
--watermark-lag-ms <MS>0Allowed lateness in milliseconds.

push flags

FlagEffect
--job-id <ID> (required)Must match a previously submitted job.
--parquet <PATH> (required)Path to a Parquet file. Rows are read into RecordBatches and pushed to the job's queue.

poll flags

FlagEffect
--job-id <ID> (required)Must match a previously submitted job.

Poll drains the accumulated output and pretty-prints it. The CLI does not checkpoint stream jobs; closing the process loses the state.

See also