Failing gracefully
•Download as PPTX, PDF•
5 likes•3,349 views
Spark Meetup (2014/09/12) の Aaron さんの資料をあずかったのでこちらにアップします。
Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (11)
Similar to Failing gracefully
Similar to Failing gracefully (20)
More from Takuya UESHIN
More from Takuya UESHIN (10)
Recently uploaded
Recently uploaded (20)
Failing gracefully
- 1. Failing Gracefully Aaron Davidson 07/01/2014
- 2. What does “failure” mean for Spark? • Spark is a cluster-compute framework targeted at analytics workloads • Supported failure modes: – Transient errors (e.g., network, HDFS outage) – Worker machine failures • Unsupported failure modes: – Systemic exceptions (e.g., bad code, OOMs) – Driver machine failure
- 3. What makes a recovery model good? • A good recovery model should: – Be simple – Consistently make progress towards completion – Always be in use (“fail constantly”)
- 4. Outline of this talk • Spark architecture overview • Common failures • Special considerations for fault tolerance
- 5. Example program Goal: Find number of names per “first character” sc.textFile(“hdfs:/names”) .map(name => (name.charAt(0), 1)) .reduceByKey(_ + _) .collect() うえださいとうえしん
- 6. Example program Goal: Find number of names per “first character” sc.textFile(“hdfs:/names”) .map(name => (name.charAt(0), 1)) .reduceByKey(_ + _) .collect() うえださいとうえしん (う, 1) (さ, 1) (う, 1)
- 7. Example program Goal: Find number of names per “first character” sc.textFile(“hdfs:/names”) .map(name => (name.charAt(0), 1)) .reduceByKey(_ + _) .collect() うえださいとうえしん (う, 1) (さ, 1) (う, 1) (う, 2) (さ, 1)
- 8. Example program Goal: Find number of names per “first character” sc.textFile(“hdfs:/names”) .map(name => (name.charAt(0), 1)) .reduceByKey(_ + _) .collect() うえださいとうえしん (う, 1) (さ, 1) (う, 1) (う, 2) (さ, 1) res0 = [(う,2), (さ,1)]
- 9. Spark Execution Model 1. Create DAG of RDDs to represent computation 2. Create logical execution plan for DAG 3. Schedule and execute individual tasks
- 10. Step 1: Create RDDs sc.textFile(“hdfs:/names”) map(name => (name.charAt(0), 1)) reduceByKey(_ + _) collect()
- 11. Step 1: Create RDDs HadoopRDD map() reduceByKey() collect()
- 12. Step 2: Create execution plan • Pipeline as much as possible • Split into “stages” based on need to reorganize data Stage 1 HadoopRDD map() reduceByKey() collect() うえださいとうえしん (う, 1) (さ, 1) (う, 1) (う, 2) (さ, 1) res0 = [(う,2), (さ,1)] Stage 2
- 13. Step 3: Schedule tasks • Split each stage into tasks • A task is data + computation • Execute all tasks within a stage before moving on
- 14. Step 3: Schedule tasks Computation Data hdfs:/names/0.gz hdfs:/names/1.gz hdfs:/names/2.gz Task 1 hdfs:/names/3.gz … Stage 1 HadoopRDD map() Task 0 hdfs:/names/0.gz HadoopRDD map() Task 1 hdfs:/names/1.gz HadoopRDD map()
- 15. Step 3: Schedule tasks /names/0.gz /names/3.gz /names/0.gz HadoopRDD map() Time HDFS HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz
- 16. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() Time
- 17. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() Time
- 18. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() Time
- 19. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() Time
- 20. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/2.gz HadoopRDD map() /names/1.gz HadoopRDD map() Time
- 21. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time
- 22. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 23. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 24. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 25. The Shuffle Stage 1 Stage 2 HadoopRDD map() reduceByKey() collect()
- 26. The Shuffle • Redistributes data among partitions • Hash keys into buckets Stage 1 Stage 2 • On reduce side, build hash map within each partition Reduce 0: { う=> 137, さ=> 86, … } Reduce 1: { な=> 144, る=> 12, … } …
- 27. The Shuffle Disk Stage 1 Stage 2 • Pull-based, not push-based • Write intermediate files to disk
- 28. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz Time /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() /names/3.gz HadoopRDD map() Reduce 0 reduceByKey collect Reduce 1 reduceByKey collect Reduce 2 reduceByKey collect Reduce 3 reduceByKey collect
- 29. When things go wrong • Task failure • Task taking a long time • Executor failure
- 30. Task Failure • Task fails with exception retry it • RDDs are immutable and “stateless”, so rerunning should have same effect – Special logic required for tasks which write data out (atomic rename) – Statelessness not enforced by programming model sc.parallelize(0 until 100).map { x => val myVal = sys.prop(“foo”, 0) + x sys.prop(“foo”) = myVal myVal }
- 31. Task Failure HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map() /names/2.gz HadoopRDD map()
- 32. Speculative Execeution • Try to predict slow or failing tasks, restart task on a different machine in parallel • Also assumes immutability and statelessness • Enable with “spark.speculation=true”
- 33. Speculative Execution HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 34. Speculative Execution HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 35. Speculative Execution HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map()
- 36. Speculative Execution HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map() /names/3.gz HadoopRDD map()
- 37. Speculative Execution HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Time /names/3.gz HadoopRDD map() /names/3.gz HadoopRDD map()
- 38. Executor Failure • Examine tasks run on that executor: – If task from final stage, we’ve already collected its results – don’t rerun – If task from intermediate stage, must rerun. • May require executing “finished” stage
- 39. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz Time /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() /names/3.gz HadoopRDD map() Reduce 0 reduceByKey collect Reduce 1 reduceByKey collect Reduce 2 reduceByKey collect Reduce 3 reduceByKey collect
- 40. Step 3: Schedule tasks HDFS /names/0.gz /names/3.gz HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz Time /names/0.gz HadoopRDD map() /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() /names/3.gz HadoopRDD map() Reduce 0 reduceByKey collect Reduce 1 reduceByKey collect Reduce 2 reduceByKey collect Reduce 3 reduceByKey collect
- 41. Step 3: Schedule tasks /names/0.gz HadoopRDD map() HDFS /names/1.gz /names/0.gz HDFS /names/2.gz /names/3.gz Time /names/1.gz HadoopRDD map() /names/2.gz HadoopRDD map() Reduce 0 reduceByKey collect Reduce 1 reduceByKey collect Reduce 2 reduceByKey collect Reduce 3 reduceByKey collect Tasks to rerun: /names/3.gz HadoopRDD map() Completed tasks: /names/0.gz HadoopRDD map() /names/3.gz HadoopRDD map() Reduce 3 reduceByKey collect
- 42. Other Failure Scenarios What happens when: 1. We have a large number of stages? 2. Our input data is not immutable (e.g. streaming)? 3. Executors had cached data?
- 43. 1. Dealing with many stages Problem: Executor loss causes recomputation of all non-final stages Solution: Checkpoint whole RDD to HDFS periodically Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Stage 6 Stage 7
- 44. 1. Dealing with many stages Problem: Executor loss causes recomputation of all non-final stages Solution: Checkpoint whole RDD to HDFS periodically HDFS Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Stage 6 Stage 7
- 45. 2. Dealing with lost input data Problem: Input data is consumed when read (e.g., streaming), and re-execution is not possible. Solution: No general solution today – either use an HDFS source or implement it yourself. Spark 1.2 roadmap includes a general solution which may trade throughput for safety.
- 46. 3. Loss of cached data Problem: Executor loss causes cache to become incomplete. Solution: Do nothing – a task caches data locally while it runs, causing the cache to stabilize.
- 47. 3. Loss of cached data val file = sc.textFile(“s3n://”).cache() // 8 blocks for (i <- 0 until 10) { file.count() } Cache Block 0 Block 2 Block 4 Block 6 Block 0 Block 2 Block 4 Block 6 Cache Block 1 Block 3 Block 5 Block 7 Block 1 Block 3 Block 5 Block 7 i = 0 i = 1 Block 0 Block 2 Block 4 Block 6 Block 1 Block 3 Block 5 Block 7
- 48. 3. Loss of cached data val file = sc.textFile(“s3n://”).cache() for (i <- 0 until 10) { file.count() } Cache i = 0 i = 1 Block 2 Block 4 Block 6 Block 7 Cache Block 1 Block 3 Block 5 Block 7 Block 1 Block 3 Block 5 Block 7 Block 1 Block 3 Block 5 Block 7 Block 2 Block 4 Block 6 Block 7 Block 0 Block 1 Block 3 Block 5 Block 0 i = 2 i = 3 Block 2 Block 4 Block 6 Block 7 Block 0 Block 1 Block 3 Block 5
- 49. Conclusions • Spark comes equipped to handle the most common forms of failure • Special care must be taken in certain cases: – Highly iterative use-cases (checkpointing) – Streaming (atomic data consumption) – Violating Spark’s core immutability and statelessness assumptions