Get the end or formerly know as "available" offsets of the latest batch that has been planned

Await until all fields of the query have been initialized.

A lock used to wait/notify when batches complete. Use a fair lock to avoid thread starvation.

Returns the path of a file with name in the checkpoint directory.

Manages the metadata from this checkpoint location.

Any clean up that needs to happen when the query is stopped or exits

Tracks how much data we have processed and committed to the sink or state store from each input source. Only the scheduler thread should modify this field, and only in atomic steps. Other threads should make a shallow copy if they are going to access this field more than once, since the field's value may change at any time.

Returns the StreamingQueryException if the query was terminated by an exception.

Expose for tests

Get the latest execution context .

Get the start offsets of the latest batch that has been planned

Interrupts the query execution thread and awaits its termination until until it exceeds the timeout. The timeout can be set on "spark.sql.streaming.stopTimeout".

Whether the query is currently active or not

The base logical plan which will be used across batch runs. Once the value is set, it should not be modified.

Called after the microbatch has completed execution. It takes care of committing the offset to commit log and other bookkeeping.

Method called once after the planning is done and before the start of the microbatch execution. It can be used to perform any pre-execution tasks.

Should not call super method as we need to do something completely different in this method for async progress tracking

A flag to indicate that a batch has completed with no new data available.

Populate the start offsets to start the execution at the current offsets stored in the sink (i.e. avoid reprocessing data that we have already processed). This function must be called before any processing occurs and will populate the following fields in the execution context of this micro-batch

• batchId
• startOffset
• endOffsets The basic structure of this method is as follows:

Identify (from the offset log) the offsets used to run the last batch IF last batch exists THEN Set the next batch to be executed as the last recovered batch Check the commit log to see which batch was committed last IF the last batch was committed THEN Call getBatch using the last batch start and end offsets // ^^^^ above line is needed since some sources assume last batch always re-executes Setup for a new batch i.e., start = last batch end, and identify new end DONE ELSE Identify a brand new batch DONE

Pretty identified string of printing in logs. Format is If name is set "queryName [id = xyz, runId = abc]" else "[id = xyz, runId = abc]"

The thread that runs the micro-batches of this stream. Note that this thread must be org.apache.spark.util.UninterruptibleThread to workaround KAFKA-1894: interrupting a running KafkaConsumer may cause endless loop.

Repeatedly attempts to run batches as data arrives.

We will only set the lastExecutionContext only if the batch id is larger than the batch id of the current latestExecutionContext. This is done to make sure we will always tracking the latest execution context i.e. we will never set latestExecutionContext to a earlier / older batch.

The list of stream instances which will be used across batch runs. Once the value is set, it should not be modified.

<invalid inheritdoc annotation>

Isolated spark session to run the batches with.

Starts the execution. This returns only after the thread has started and QueryStartedEvent has been posted to all the listeners.

Defines the internal state of execution

Signals to the thread executing micro-batches that it should stop running after the next batch. This method blocks until the thread stops running.

Stops all streaming sources safely.

Used to report metrics to coda-hale. This uses id for easier tracking across restarts.

A list of unique sources in the query plan. This will be set when generating logical plan.

Conduct sanity checks on the offset log to make sure it is correct and expected. Also return the previous offset written to the offset log

A map of current watermarks, keyed by the position of the watermark operator in the physical plan.

This state is 'soft state', which does not affect the correctness and semantics of watermarks and is not persisted across query restarts. The fault-tolerant watermark state is in offsetSeqMetadata.