> On Sept. 12, 2016, 6:26 p.m., Joshua Cohen wrote:
> > Along the lines of the question Stephan raised, what happens in the event 
> > of a failover mid-batch, especially w.r.t. repeatable work?

Same as today: the transaction either happens or does not. In case of a cron 
job (the only user of `RepeatableWork` at the moment), the cron job will not be 
triggered (again, same as today).


> On Sept. 12, 2016, 6:26 p.m., Joshua Cohen wrote:
> > src/main/java/org/apache/aurora/scheduler/BatchWorker.java, lines 106-107
> > <https://reviews.apache.org/r/51759/diff/1/?file=1495031#file1495031line106>
> >
> >     I think this should be "does not initialize"?

Good catch. Fixed.


> On Sept. 12, 2016, 6:26 p.m., Joshua Cohen wrote:
> > src/main/java/org/apache/aurora/scheduler/BatchWorker.java, lines 257-259
> > <https://reviews.apache.org/r/51759/diff/1/?file=1495031#file1495031line257>
> >
> >     Even though there's only a single callsite where we initialize a 
> > `WorkItem` in `FOLLOWUP` state, there's no guarantee that `laskBackoffMsec` 
> > will be present just because the state is `FOLLOWUP`.
> >     
> >     We should add some state checks to the `WorkItem` constructor so that 
> > if state is `FOLLOWUP` we'll throw if `lastBackoffMsec` is not present to 
> > protect against future mistakes? (Or if that's a valid scenario we should 
> > perform an `isPresent` check or use `or` here...).

I assert both `backoffStrategy` and `lastBackoffMsec` inside the `backoffFor()` 
method on line 315. I think it should be sufficient given that `WorkItem` is a 
private class?


> On Sept. 12, 2016, 6:26 p.m., Joshua Cohen wrote:
> > src/main/java/org/apache/aurora/scheduler/BatchWorker.java, lines 273-275
> > <https://reviews.apache.org/r/51759/diff/1/?file=1495031#file1495031line273>
> >
> >     Should we do this after the batch is processed rather than before?

It's largely irrelevant but I guess it's more logical to expect it at the end 
while going through the code. Moved.


- Maxim


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/51759/#review148501
-----------------------------------------------------------


On Sept. 9, 2016, 5:29 p.m., Maxim Khutornenko wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/51759/
> -----------------------------------------------------------
> 
> (Updated Sept. 9, 2016, 5:29 p.m.)
> 
> 
> Review request for Aurora, Joshua Cohen, Stephan Erb, and Zameer Manji.
> 
> 
> Repository: aurora
> 
> 
> Description
> -------
> 
> This is the first (out of 3) patches intending to reduce storage write lock 
> contention and as such improve overall system write throughput. It introduces 
> the `BatchWorker` and migrates the majority of storage writes due to task 
> status change events to use `TaskEventBatchWorker`.
> 
> #####Problem
> Our current storage system writes effectively behave as `SERIALIZABLE` 
> transaction isolation level in SQL terms. This means all writes require 
> exclusive access to the storage and no two transactions can happen in 
> parallel [1]. While it certainly simplifies our implementation, it creates a 
> single hotspot where multiple threads are competing for the storage write 
> access. This type of contention only worsens as the cluster size grows, more 
> tasks are scheduled, more status updates are processed, more subscribers are 
> listening to status updates and etc. Eventually, the scheduler throughput 
> (and especially task scheduling) becomes degraded to the extent that certain 
> operations wait much longer (4x and more) for the lock acquisition than it 
> takes to process their payload when inside the transaction. Some ops (like 
> event processing) are generally tolerant of these types of delays. Others - 
> not as much. The task scheduling suffers the most as backing up the 
> scheduling queue directly affects
  the Median Time To Assigned (MTTA).
> 
> #####Remediation
> Given the above, it's natural to assume that reducing the number of write 
> transactions should help reducing the lock contention. This patch introduces 
> a generic `BatchWorker` service that delivers a "best effort" batching 
> approach by redirecting multiple individual write requests into a single FIFO 
> queue served non-stop by a single dedicated thread. Every batch shares a 
> single write transaction thus reducing the number of potential write lock 
> requests. To minimize wait-in-queue time, items are dispatched immediately 
> and the max number of items is bounded. There are a few `BatchWorker` 
> instances specialized on particular workload types: task even processing, 
> cron scheduling and task scheduling. Every instance can be tuned 
> independently (max batch size) and provides specialized metrics helping to 
> monitor each workload type perf.
> 
> #####Results
> The proposed approach has been heavily tested in production and delivered the 
> best results. The lock contention latencies got down between 2x and 5x 
> depending on the cluster load. A number of other approaches tried but 
> discarded as not performing well or even performing much worse than the 
> current master:
> - Clock-driven batch execution - every batch is dispatched on a time schedule
> - Max batch with a deadline - a batch is dispatched when max size is reached 
> OR a timeout expires
> - Various combinations of the above - some `BatchWorkers` are using 
> clock-driven execution while others are using max batch with a deadline
> - Completely non-blocking (event-based) completion notification - all call 
> sites are notified of item completion via a `BatchWorkCompleted` event
> 
> Happy to provide more details on the above if interested.
> 
> #####Upcoming
> The introduction of the `BatchWorker` by itself was not enough to 
> substantially improve the MTTA. It, however, paves the way for the next phase 
> of scheduling perf improvement - taking more than 1 task from a given 
> `TaskGroup` in a single scheduling round (coming soon). That improvement 
> wouldn't deliver without decreasing the lock contention first. 
> 
> Note: it wasn't easy to have a clean diff split, so some functionality in 
> `BatchWorker` (e.g.: `executeWithReplay`) appears to be unused in the current 
> patch but will become obvious in the part 2 (coming out shortly).  
> 
> [1] - 
> https://github.com/apache/aurora/blob/f6ac13b169aaad5aad73ef3cc72873781e30a705/src/main/java/org/apache/aurora/scheduler/storage/log/LogStorage.java#L540-L555
> 
> 
> Diffs
> -----
> 
>   src/main/java/org/apache/aurora/scheduler/BatchWorker.java PRE-CREATION 
>   src/main/java/org/apache/aurora/scheduler/SchedulerModule.java 
> 4a7ef0b1b90607f68d89fe8e207f42c42a8c56a0 
>   src/main/java/org/apache/aurora/scheduler/events/PubsubEvent.java 
> 70b5470b9dad1af838b5222cae5ac86487e2f2e4 
>   src/main/java/org/apache/aurora/scheduler/pruning/TaskHistoryPruner.java 
> f07746c2b990c1c2235e99f9e4775fc84f9c27b1 
>   src/main/java/org/apache/aurora/scheduler/scheduling/TaskThrottler.java 
> bbd971a2aa8a96cf79edd879ad60e1bebd933d79 
>   src/main/java/org/apache/aurora/scheduler/state/MaintenanceController.java 
> 3c7cda09ab292d696070ca4d9dfedb1f6f71b0fe 
>   
> src/main/java/org/apache/aurora/scheduler/updater/JobUpdateControllerImpl.java
>  594bb6219294dcc77d48dcad14e2a6f9caa0c534 
>   src/test/java/org/apache/aurora/scheduler/BatchWorkerTest.java PRE-CREATION 
>   
> src/test/java/org/apache/aurora/scheduler/pruning/TaskHistoryPrunerTest.java 
> 99c27e8012f10a67ce5f1b84d258e7a5608995c7 
>   src/test/java/org/apache/aurora/scheduler/scheduling/TaskThrottlerTest.java 
> 7d104aa2ea4a4d99be4711f666d18beca238284e 
>   
> src/test/java/org/apache/aurora/scheduler/state/MaintenanceControllerImplTest.java
>  94f5ca565476f62d72879837a0e7dafabcf30432 
>   src/test/java/org/apache/aurora/scheduler/testing/BatchWorkerUtil.java 
> PRE-CREATION 
>   src/test/java/org/apache/aurora/scheduler/updater/JobUpdaterIT.java 
> 196df4754b553f05e50b66ad2f84271901bc9eba 
> 
> Diff: https://reviews.apache.org/r/51759/diff/
> 
> 
> Testing
> -------
> 
> All types of testing including deploying to test and production clusters.
> 
> 
> Thanks,
> 
> Maxim Khutornenko
> 
>

Reply via email to