To bring consistency to docs and source this commit changes the default
values for maxRowsInMemory and rowFlushBoundary to 75000 after
discussion in PR https://github.com/druid-io/druid/pull/2457.
The previous default was 500000 and it's lower now on the grounds that
it's better for a default to be somewhat less efficient, and work,
than to reach for the stars and possibly result in
"OutOfMemoryError: java heap space" errors.
- Add TaskLocation class
- Add registerListener to TaskRunner
- Add getLocation to TaskRunnerWorkItem
- Implement location tracking in existing TaskRunners
- Rework WorkerTaskMonitor to do management out of a single thread so it can
handle status and location updates more simply.
* Moves last run task state information to Worker
* Makes WorkerTaskRunner a TaskRunner which has interfaces to help with getting information about a Worker
Two changes:
- Allow IncrementalIndex to suppress ParseExceptions on "aggregate".
- Add "reportParseExceptions" option to realtime tuning configs. By default this is "false".
Behavior of the counters should now be:
- processed: Number of rows indexed, including rows where some fields could be parsed and some could not.
- thrownAway: Number of rows thrown away due to rejection policy.
- unparseable: Number of rows thrown away due to being completely unparseable (no fields salvageable at all).
If "reportParseExceptions" is true then "unparseable" will always be zero (because a parse error would
cause an exception to be thrown). In addition, "processed" will only include fully parseable rows
(because even partial parse failures will cause exceptions to be thrown).
Fixes#2510.
* Don't put druid****selfcontained.jar at the end of the hadoop isolated classpath
* Add `<scope>provided</scope>` to prevent repeated dependency inclusion in the extension directories
- Throw most exceptions rather than suppressing them, which should help
detect problems. Continue suppressing exceptions that make sense to
suppress.
- Handle payload length checks consistently, and improve error message.
- Remove unused WorkerCuratorCoordinator.announceTaskAnnouncement method.
- Max znode length should be int, not long.
- Add tests.
* Defaults the thread priority to java.util.Thread.NORM_PRIORITY in io.druid.indexing.common.task.AbstractTask
* Each exec service has its own Task Factory which is assigned a priority for spawned task. Therefore each priority class has a unique exec service
* Added priority to tasks as taskPriority in the task context. <0 means low, 0 means take default, >0 means high. It is up to any particular implementation to determine how to handle these numbers
* Add options to ForkingTaskRunner
* Add "-XX:+UseThreadPriorities" default option
* Add "-XX:ThreadPriorityPolicy=42" default option
* AbstractTask - Removed unneded @JsonIgnore on priority
* Added priority to RealtimePlumber executors. All sub-executors (non query runners) get Thread.MIN_PRIORITY
* Add persistThreadPriority and mergeThreadPriority to realtime tuning config
Historical will drop a segment that shouldn't be dropped in the following scenario:
Historical node tried to load segmentA, but failed with SegmentLoadingException,
then ZkCoordinator called removeSegment(segmentA, blah) to schedule a runnable that would drop segmentA by deleting its files. Now, before that runnable executed, another LOAD request was sent to this historical, this time historical actually succeeded on loading segmentA and announced it. But later on, the scheduled drop-of-segment runnable started executing and removed the segment files, while historical is still announcing segmentA.
- Shedding locks at startup is bad, we actually want to keep them. Stop doing that.
- stopGracefully now interrupts the run thread if had started running finishJob. This avoids
waiting for handoff unnecessarily.
- fixes#1970
- extracted out segment handoff callbacks in SegmentHandoffNotifier
which is responsible for tracking segment handoffs and doing callbacks
when handoff is complete.
- Coordinator now maintains a view of segments in the cluster, this
will affect the jam heap requirements for the overlord for large
clusters.
realtime index task and nodes now use HTTP end points exposed by the
coordinator to get serverView
review comment
fix realtime node guide injection
review comments
make test not rely on scheduled exec
fix compilation
fix import
review comment
introduce immutableSegmentLoadInfo
fix son reading
remove unnecessary logging
This is done by killing and respawning the jvms rather than reconnecting to existing
jvms, for a couple reasons. One is that it lets you restore tasks after server reboots
too, and another is that it lets you upgrade all the software on a box at once by just
restarting everything.
The main changes are,
1) Add "canRestore" and "stopGracefully" methods to Tasks that say if a task can
stop gracefully, and actually do a graceful stop. RealtimeIndexTask is the only
one that currently implements this.
2) Add "stop" method to TaskRunners that attempts to do an orderly shutdown.
ThreadPoolTaskRunner- call stopGracefully on restorable tasks, wait for exit
ForkingTaskRunner- close output stream to restorable tasks, wait for exit
RemoteTaskRunner- do nothing special, we actually don't want to shutdown
3) Add "restore" method to TaskRunners that attempts to bootstrap tasks from last run.
Only ForkingTaskRunner does anything here. It maintains a "restore.json" file with
a list of restorable tasks.
4) Have the CliPeon's ExecutorLifecycle lock the task base directory to avoid a restored
task and a zombie old task from stomping on each other.
Deserialization of Optionals does not work quite right- they come back as actual
nulls, rather than absent Optionals. So these probably only ever worked for the local
task action client.
This is a feature meant to allow realtime tasks to work without being told upfront
what shardSpec they should use (so we can potentially publish a variable number
of segments per interval).
The idea is that there is a "pendingSegments" table in the metadata store that
tracks allocated segments. Each one has a segment id (the same segment id we know
and love) and is also part of a sequence.
The sequences are an idea from @cheddar that offers a way of doing replication.
If there are N tasks reading exactly the same data with exactly the same logic
(think Kafka tasks reading a fixed range of offsets) then you can place them
in the same sequence, and they will generate the same sequence of segments.
In #933 the ForkingTaskRunner's logging was changed to buffered from
unbuffered. This means that the last few KB of the logs are generally
not visible while a task is running, which makes debugging running
tasks difficult.
Gian Merlino
Björn Zettergren
Charles Allen
Himanshu Gupta
Fangjin Yang
Nishant
Bingkun Guo
Andrés Gomez
Xavier Léauté
Parag Jain
Charles Allen
Jonathan Wei
Kurt Young
navis.ryu
Xavier Léauté
Bartosz Ługowski
zhxiaog