Skip to main content
Version: 0.15

Event time and watermark

Event time is a timestamp attached to a row. The timestamp can be anything but it is usually the time the row is generated. A sensor might add a timestamp to a sensor data, and stream processors can use the timestamp as an event time.

Event time is different from the real time (wall-clock time). Stream processors may input rows from log files that have lines with past timestamps. The past timestamps can be dealt with event times.

Watermark is similar to a clock in event-time world. A watermark usually proceeds when a stream processor observes a row with the latest event time.

Watermark is used to deal with late data. A stream processor may observe 12:09:59 (event time) after 12:10:00. Watermark provides a criteria for users of stream processors to decide how to deal with the 12:09:59 row.

We will explain them in the future but currently you can refer to the following materials to get understand them:

Event time in SpringQL​

In SpringQL, you can specify which column to regard as an event time in stream definitions.

Defining an event time column
CREATE STREAM s (
  c1 TIMESTAMP NOT NULL ROWTIME,
  c2 TIMESTAMP NOT NULL
);

c1 is a TIMESTAMP type and followed by ROWTIME keyword to describe the column is the event time in the stream s. c2 also has the TIMESTAMP type but it is not an event time because it does not have ROWTIME keyword.

Watermark in SpringQL​

SpringQL has a watermark for a window. If you create a pump with a WINDOW clause, you implicitly create a watermark.

A watermark is locally used in a window to point the current time in event-time.

Currently, SpringQL adopts only a simple way to update a watermark: A watermark is the maximum event time a window observes.

  1. A window is generated from CREATE PUMP ... WINDOW ...;

    -> Watermark is undefined

  2. The window inputs 1st row with 12:00:00 (event time).

    -> Watermark is 12:00:00

  3. The window inputs 2nd row with 12:01:00.

    -> Watermark is 12:01:00

  4. The window inputs 2nd row with 12:00:30 (late data).

    -> Watermark is still 12:01:00