[SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors #306
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Can one of the admins verify this patch? |
| @@ -142,7 +142,7 @@ private[spark] class ExecutorRunner( | |||
| // long-lived processes only. However, in the future, we might restart the executor a few | |||
| // times on the same machine. | |||
| val exitCode = process.waitFor() | |||
| state = ExecutorState.FAILED | |||
There was a problem hiding this comment.
so, when will you use FAILED? I think at least this place, it should be FAILED...because the executor is actually stopped by InterruptException, if your execution flow comes to this place, it means it is certainly failed
There was a problem hiding this comment.
BTW, do you mind fixing the comments... I think it's out of date according to your finding......the user program can call sc.stop() to explicitly exit the Executor now, and the return value is 0....
There was a problem hiding this comment.
Sure, I'll update the comments.
There was a problem hiding this comment.
BTW, when Worker kills an executor, the ExecutorRunner.workerThread got interrupted, while the Executor process itself is destroyed. The above line in ExecutorRunner will not get executed, since the thread got interrupted. The executor state will not be set here. Instead, it is set in the ExecutorRunner.kill() as KILLED.
There was a problem hiding this comment.
yes, that's what I was saying, I thought this line would never be executed in normal case, once executed, it should be FAILED, (I missed the sc.stop() calling )
|
When the driver sends StopExecutor to executor (log msg "Driver commanded a shutdown"), it also ends up here. So it can be a normal exit depends on the exit code. See description. There are 2 other places where FAILED is used. One is when exception is caught in this same method. The other is when exception is caught in LaunchExecutor() in Worker.scala. |
|
so, you are assuming that the user always explicitly call sc.stop(), but that's not always true, even the example program in spark's document does not do that http://spark.apache.org/docs/latest/quick-start.html#a-standalone-app-in-scala |
|
That's ok. If the driver doesn't instruct executor to shutdown, the process won't exit, right? The executor will eventually be KILLED by Master when app disassociates. |
|
@kanzhang I think the tricky part is just here, a normally exited program seems the same with the ones being interrupted during the running .....though I think your PR can solve the program and in the other hand, we can remind user to call sc.stop() after everything is done in 1.0 document..... |
|
Thanks for your comments. I just did a test of the ctrl-c case using Spark Shell. When I used ctrl-c to exit the shell, both Executor log and Master log showed that they detected app/driver disassociated. The Executor tries to shut itself down with exit code 1. However, the Master's finishApplication() wins in my test. Worker's log shows it accepted Master's request to kill the executor and the executor's final state is KILLED. It could have happened the other way with the executor exited with code 1, Worker marked its state as EXITED, and Master tried to launch a replacement executor (since exit code is not 0), which eventually got KILLED when Master detects app disassociated. Either way is fine in my opinion. I don't think we need to force users to call stop(). It's just that if it is not called, its executors may get KILLED. |
|
Code comments updated |
…exited executors updated code comments minor comment wording change
kanzhang
changed the title
|
Since SPARK-1118 has been closed as duplicate of SPARK-937, I'm re-submitting this PR as SPARK-937. @aarondav feel free to take over if you still plan to work on SPARK-937. |
|
@aarondav can you take a look again at this? |
|
I agree that this may not cause the status to be EXITED in all cases, but this definitely seems like an improvement over the prior behavior. One thing I noticed while testing this PR is that if you do exit cleanly, the Master Web UI now actually forgets about all the Executors! So you wouldn't see the "EXITED" status at all. I created SPARK-2147 to track this issue. |
|
Jenkins, test this please. |
|
Merged build triggered. |
|
Merged build started. |
|
Merged build finished. All automated tests passed. |
|
All automated tests passed. |
|
@aarondav thanks for reviewing this patch and tracking remaining issues. Pls let me know if there is anything else I can do to improve this patch. |
|
Cool, merged into master and branch-1.0. Thanks! |
asfgit
pushed a commit
that referenced
this pull request
Jun 15, 2014
…exited executors There seems to be 2 issues. 1. When job is done, driver asks executor to shutdown. However, this clean exit was assigned FAILED executor state by Worker. I introduced EXITED executor state for executors who voluntarily exit (both normal and abnormal exit depending on the exit code). 2. When Master gets notified an executor has exited, it launches another one to replace it, regardless of reason why the executor had exited. When the reason was job has finished, the unnecessary replacement got subsequently killed when App disassociates. This launching and killing of unnecessary executors shows up in the log and is confusing to users. I added check for executor exit status and avoid launching (and subsequent killing) of unnecessary replacements when executors exit cleanly. One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements. Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED. Author: Kan Zhang <kzhang@apache.org> Closes #306 from kanzhang/SPARK-1118 and squashes the following commits: cb0cc86 [Kan Zhang] [SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors (cherry picked from commit ca5d9d4) Signed-off-by: Aaron Davidson <aaron@databricks.com>
pdeyhim
pushed a commit
to pdeyhim/spark-1
that referenced
this pull request
Jun 25, 2014
…exited executors There seems to be 2 issues. 1. When job is done, driver asks executor to shutdown. However, this clean exit was assigned FAILED executor state by Worker. I introduced EXITED executor state for executors who voluntarily exit (both normal and abnormal exit depending on the exit code). 2. When Master gets notified an executor has exited, it launches another one to replace it, regardless of reason why the executor had exited. When the reason was job has finished, the unnecessary replacement got subsequently killed when App disassociates. This launching and killing of unnecessary executors shows up in the log and is confusing to users. I added check for executor exit status and avoid launching (and subsequent killing) of unnecessary replacements when executors exit cleanly. One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements. Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED. Author: Kan Zhang <kzhang@apache.org> Closes apache#306 from kanzhang/SPARK-1118 and squashes the following commits: cb0cc86 [Kan Zhang] [SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors
markhamstra
pushed a commit
to markhamstra/spark
that referenced
this pull request
Jul 10, 2014
…exited executors There seems to be 2 issues. 1. When job is done, driver asks executor to shutdown. However, this clean exit was assigned FAILED executor state by Worker. I introduced EXITED executor state for executors who voluntarily exit (both normal and abnormal exit depending on the exit code). 2. When Master gets notified an executor has exited, it launches another one to replace it, regardless of reason why the executor had exited. When the reason was job has finished, the unnecessary replacement got subsequently killed when App disassociates. This launching and killing of unnecessary executors shows up in the log and is confusing to users. I added check for executor exit status and avoid launching (and subsequent killing) of unnecessary replacements when executors exit cleanly. One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements. Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED. Author: Kan Zhang <kzhang@apache.org> Closes apache#306 from kanzhang/SPARK-1118 and squashes the following commits: cb0cc86 [Kan Zhang] [SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors Conflicts: core/src/main/scala/org/apache/spark/deploy/master/Master.scala core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
xiliu82
pushed a commit
to xiliu82/spark
that referenced
this pull request
Sep 4, 2014
…exited executors There seems to be 2 issues. 1. When job is done, driver asks executor to shutdown. However, this clean exit was assigned FAILED executor state by Worker. I introduced EXITED executor state for executors who voluntarily exit (both normal and abnormal exit depending on the exit code). 2. When Master gets notified an executor has exited, it launches another one to replace it, regardless of reason why the executor had exited. When the reason was job has finished, the unnecessary replacement got subsequently killed when App disassociates. This launching and killing of unnecessary executors shows up in the log and is confusing to users. I added check for executor exit status and avoid launching (and subsequent killing) of unnecessary replacements when executors exit cleanly. One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements. Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED. Author: Kan Zhang <kzhang@apache.org> Closes apache#306 from kanzhang/SPARK-1118 and squashes the following commits: cb0cc86 [Kan Zhang] [SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors
bzhaoopenstack
pushed a commit
to bzhaoopenstack/spark
that referenced
this pull request
Sep 11, 2019
Remove uselesse jobs
cloud-fan
pushed a commit
that referenced
this pull request
Jan 3, 2020
### What changes were proposed in this pull request? It is very common for a SQL query to query a table more than once. For example: ``` == Physical Plan == *(12) HashAggregate(keys=[cmn_mtrc_summ_dt#21, rev_rollup#1279, CASE WHEN (rev_rollup#1319 = rev_rollup#1279) THEN 0 ELSE 1 END#1366, CASE WHEN cast(sap_category_id#24 as decimal(10,0)) IN (5,7,23,41) THEN 0 ELSE 1 END#1367], functions=[sum(coalesce(bid_count#34, 0)), sum(coalesce(ck_trans_count#35, 0)), sum(coalesce(ended_bid_count#36, 0)), sum(coalesce(ended_lstg_count#37, 0)), sum(coalesce(ended_success_lstg_count#38, 0)), sum(coalesce(item_sold_count#39, 0)), sum(coalesce(new_lstg_count#40, 0)), sum(coalesce(gmv_us_amt#41, 0.00)), sum(coalesce(gmv_slr_lc_amt#42, 0.00)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_insrtn_fee_us_amt#46, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_insrtn_crd_us_amt#50, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_fetr_fee_us_amt#54, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_fetr_crd_us_amt#58, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_fv_fee_us_amt#62, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_fv_crd_us_amt#67, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_othr_l_fee_us_amt#72, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_othr_l_crd_us_amt#76, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_othr_nl_fee_us_amt#80, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_othr_nl_crd_us_amt#84, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_slr_tools_fee_us_amt#88, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_slr_tools_crd_us_amt#92, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), sum(coalesce(rvnu_unasgnd_us_amt#96, 0.000000)), sum((coalesce(rvnu_transaction_us_amt#112, 0.0) + coalesce(rvnu_transaction_crd_us_amt#115, 0.0))), sum((coalesce(rvnu_total_us_amt#118, 0.0) + coalesce(rvnu_total_crd_us_amt#121, 0.0)))]) +- Exchange hashpartitioning(cmn_mtrc_summ_dt#21, rev_rollup#1279, CASE WHEN (rev_rollup#1319 = rev_rollup#1279) THEN 0 ELSE 1 END#1366, CASE WHEN cast(sap_category_id#24 as decimal(10,0)) IN (5,7,23,41) THEN 0 ELSE 1 END#1367, 200), true, [id=#403] +- *(11) HashAggregate(keys=[cmn_mtrc_summ_dt#21, rev_rollup#1279, CASE WHEN (rev_rollup#1319 = rev_rollup#1279) THEN 0 ELSE 1 END AS CASE WHEN (rev_rollup#1319 = rev_rollup#1279) THEN 0 ELSE 1 END#1366, CASE WHEN cast(sap_category_id#24 as decimal(10,0)) IN (5,7,23,41) THEN 0 ELSE 1 END AS CASE WHEN cast(sap_category_id#24 as decimal(10,0)) IN (5,7,23,41) THEN 0 ELSE 1 END#1367], functions=[partial_sum(coalesce(bid_count#34, 0)), partial_sum(coalesce(ck_trans_count#35, 0)), partial_sum(coalesce(ended_bid_count#36, 0)), partial_sum(coalesce(ended_lstg_count#37, 0)), partial_sum(coalesce(ended_success_lstg_count#38, 0)), partial_sum(coalesce(item_sold_count#39, 0)), partial_sum(coalesce(new_lstg_count#40, 0)), partial_sum(coalesce(gmv_us_amt#41, 0.00)), partial_sum(coalesce(gmv_slr_lc_amt#42, 0.00)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_insrtn_fee_us_amt#46, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_insrtn_crd_us_amt#50, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_fetr_fee_us_amt#54, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_fetr_crd_us_amt#58, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_fv_fee_us_amt#62, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_fv_crd_us_amt#67, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_othr_l_fee_us_amt#72, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_othr_l_crd_us_amt#76, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_othr_nl_fee_us_amt#80, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_othr_nl_crd_us_amt#84, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(CheckOverflow((promote_precision(cast(coalesce(rvnu_slr_tools_fee_us_amt#88, 0.000000) as decimal(19,6))) + promote_precision(cast(coalesce(rvnu_slr_tools_crd_us_amt#92, 0.000000) as decimal(19,6)))), DecimalType(19,6), true)), partial_sum(coalesce(rvnu_unasgnd_us_amt#96, 0.000000)), partial_sum((coalesce(rvnu_transaction_us_amt#112, 0.0) + coalesce(rvnu_transaction_crd_us_amt#115, 0.0))), partial_sum((coalesce(rvnu_total_us_amt#118, 0.0) + coalesce(rvnu_total_crd_us_amt#121, 0.0)))]) +- *(11) Project [cmn_mtrc_summ_dt#21, sap_category_id#24, bid_count#34, ck_trans_count#35, ended_bid_count#36, ended_lstg_count#37, ended_success_lstg_count#38, item_sold_count#39, new_lstg_count#40, gmv_us_amt#41, gmv_slr_lc_amt#42, rvnu_insrtn_fee_us_amt#46, rvnu_insrtn_crd_us_amt#50, rvnu_fetr_fee_us_amt#54, rvnu_fetr_crd_us_amt#58, rvnu_fv_fee_us_amt#62, rvnu_fv_crd_us_amt#67, rvnu_othr_l_fee_us_amt#72, rvnu_othr_l_crd_us_amt#76, rvnu_othr_nl_fee_us_amt#80, rvnu_othr_nl_crd_us_amt#84, rvnu_slr_tools_fee_us_amt#88, rvnu_slr_tools_crd_us_amt#92, rvnu_unasgnd_us_amt#96, ... 6 more fields] +- *(11) BroadcastHashJoin [byr_cntry_id#23], [cntry_id#1309], LeftOuter, BuildRight :- *(11) Project [cmn_mtrc_summ_dt#21, byr_cntry_id#23, sap_category_id#24, bid_count#34, ck_trans_count#35, ended_bid_count#36, ended_lstg_count#37, ended_success_lstg_count#38, item_sold_count#39, new_lstg_count#40, gmv_us_amt#41, gmv_slr_lc_amt#42, rvnu_insrtn_fee_us_amt#46, rvnu_insrtn_crd_us_amt#50, rvnu_fetr_fee_us_amt#54, rvnu_fetr_crd_us_amt#58, rvnu_fv_fee_us_amt#62, rvnu_fv_crd_us_amt#67, rvnu_othr_l_fee_us_amt#72, rvnu_othr_l_crd_us_amt#76, rvnu_othr_nl_fee_us_amt#80, rvnu_othr_nl_crd_us_amt#84, rvnu_slr_tools_fee_us_amt#88, rvnu_slr_tools_crd_us_amt#92, ... 6 more fields] : +- *(11) BroadcastHashJoin [slr_cntry_id#28], [cntry_id#1269], LeftOuter, BuildRight : :- *(11) Project [gen_attr_1#360 AS cmn_mtrc_summ_dt#21, gen_attr_5#267 AS byr_cntry_id#23, gen_attr_7#268 AS sap_category_id#24, gen_attr_15#272 AS slr_cntry_id#28, gen_attr_27#278 AS bid_count#34, gen_attr_29#279 AS ck_trans_count#35, gen_attr_31#280 AS ended_bid_count#36, gen_attr_33#282 AS ended_lstg_count#37, gen_attr_35#283 AS ended_success_lstg_count#38, gen_attr_37#284 AS item_sold_count#39, gen_attr_39#281 AS new_lstg_count#40, gen_attr_41#285 AS gmv_us_amt#41, gen_attr_43#287 AS gmv_slr_lc_amt#42, gen_attr_51#290 AS rvnu_insrtn_fee_us_amt#46, gen_attr_59#294 AS rvnu_insrtn_crd_us_amt#50, gen_attr_67#298 AS rvnu_fetr_fee_us_amt#54, gen_attr_75#302 AS rvnu_fetr_crd_us_amt#58, gen_attr_83#306 AS rvnu_fv_fee_us_amt#62, gen_attr_93#311 AS rvnu_fv_crd_us_amt#67, gen_attr_103#316 AS rvnu_othr_l_fee_us_amt#72, gen_attr_111#320 AS rvnu_othr_l_crd_us_amt#76, gen_attr_119#324 AS rvnu_othr_nl_fee_us_amt#80, gen_attr_127#328 AS rvnu_othr_nl_crd_us_amt#84, gen_attr_135#332 AS rvnu_slr_tools_fee_us_amt#88, ... 6 more fields] : : +- *(11) BroadcastHashJoin [cast(gen_attr_308#777 as decimal(20,0))], [cast(gen_attr_309#803 as decimal(20,0))], LeftOuter, BuildRight : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 6 more fields] : : : +- *(11) BroadcastHashJoin [cast(gen_attr_310#674 as int)], [cast(gen_attr_311#774 as int)], LeftOuter, BuildRight : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 6 more fields] : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_5#267 as decimal(20,0))], [cast(gen_attr_312#665 as decimal(20,0))], LeftOuter, BuildRight : : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 5 more fields] : : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_313#565 as decimal(20,0))], [cast(gen_attr_314#591 as decimal(20,0))], LeftOuter, BuildRight : : : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 6 more fields] : : : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_315#462 as int)], [cast(gen_attr_316#562 as int)], LeftOuter, BuildRight : : : : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 6 more fields] : : : : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_15#272 as decimal(20,0))], [cast(gen_attr_317#453 as decimal(20,0))], LeftOuter, BuildRight : : : : : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, gen_attr_143#336, ... 5 more fields] : : : : : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_25#277 as decimal(20,0))], [cast(gen_attr_318#379 as decimal(20,0))], LeftOuter, BuildRight : : : : : : : : :- *(11) Project [gen_attr_5#267, gen_attr_7#268, gen_attr_15#272, gen_attr_25#277, gen_attr_27#278, gen_attr_29#279, gen_attr_31#280, gen_attr_39#281, gen_attr_33#282, gen_attr_35#283, gen_attr_37#284, gen_attr_41#285, gen_attr_43#287, gen_attr_51#290, gen_attr_59#294, gen_attr_67#298, gen_attr_75#302, gen_attr_83#306, gen_attr_93#311, gen_attr_103#316, gen_attr_111#320, gen_attr_119#324, gen_attr_127#328, gen_attr_135#332, ... 6 more fields] : : : : : : : : : +- *(11) BroadcastHashJoin [cast(gen_attr_23#276 as decimal(20,0))], [cast(gen_attr_319#367 as decimal(20,0))], LeftOuter, BuildRight : : : : : : : : : :- *(11) Project [byr_cntry_id#1169 AS gen_attr_5#267, sap_category_id#1170 AS gen_attr_7#268, slr_cntry_id#1174 AS gen_attr_15#272, lstg_curncy_id#1178 AS gen_attr_23#276, blng_curncy_id#1179 AS gen_attr_25#277, bid_count#1180 AS gen_attr_27#278, ck_trans_count#1181 AS gen_attr_29#279, ended_bid_count#1182 AS gen_attr_31#280, new_lstg_count#1183 AS gen_attr_39#281, ended_lstg_count#1184 AS gen_attr_33#282, ended_success_lstg_count#1185 AS gen_attr_35#283, item_sold_count#1186 AS gen_attr_37#284, gmv_us_amt#1187 AS gen_attr_41#285, gmv_slr_lc_amt#1189 AS gen_attr_43#287, rvnu_insrtn_fee_us_amt#1192 AS gen_attr_51#290, rvnu_insrtn_crd_us_amt#1196 AS gen_attr_59#294, rvnu_fetr_fee_us_amt#1200 AS gen_attr_67#298, rvnu_fetr_crd_us_amt#1204 AS gen_attr_75#302, rvnu_fv_fee_us_amt#1208 AS gen_attr_83#306, rvnu_fv_crd_us_amt#1213 AS gen_attr_93#311, rvnu_othr_l_fee_us_amt#1218 AS gen_attr_103#316, rvnu_othr_l_crd_us_amt#1222 AS gen_attr_111#320, rvnu_othr_nl_fee_us_amt#1226 AS gen_attr_119#324, rvnu_othr_nl_crd_us_amt#1230 AS gen_attr_127#328, ... 7 more fields] : : : : : : : : : : +- *(11) ColumnarToRow : : : : : : : : : : +- FileScan parquet default.big_table1[byr_cntry_id#1169,sap_category_id#1170,slr_cntry_id#1174,lstg_curncy_id#1178,blng_curncy_id#1179,bid_count#1180,ck_trans_count#1181,ended_bid_count#1182,new_lstg_count#1183,ended_lstg_count#1184,ended_success_lstg_count#1185,item_sold_count#1186,gmv_us_amt#1187,gmv_slr_lc_amt#1189,rvnu_insrtn_fee_us_amt#1192,rvnu_insrtn_crd_us_amt#1196,rvnu_fetr_fee_us_amt#1200,rvnu_fetr_crd_us_amt#1204,rvnu_fv_fee_us_amt#1208,rvnu_fv_crd_us_amt#1213,rvnu_othr_l_fee_us_amt#1218,rvnu_othr_l_crd_us_amt#1222,rvnu_othr_nl_fee_us_amt#1226,rvnu_othr_nl_crd_us_amt#1230,... 7 more fields] Batched: true, DataFilters: [], Format: Parquet, Location: PrunedInMemoryFileIndex[], PartitionFilters: [isnotnull(cmn_mtrc_summ_dt#1262), (cmn_mtrc_summ_dt#1262 >= 18078), (cmn_mtrc_summ_dt#1262 <= 18..., PushedFilters: [], ReadSchema: struct<byr_cntry_id:decimal(4,0),sap_category_id:decimal(9,0),slr_cntry_id:decimal(4,0),lstg_curn... : : : : : : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(9,0), true] as decimal(20,0)))), [id=#288] : : : : : : : : : +- *(1) Project [CURNCY_ID#1263 AS gen_attr_319#367] : : : : : : : : : +- *(1) Filter isnotnull(CURNCY_ID#1263) : : : : : : : : : +- *(1) ColumnarToRow : : : : : : : : : +- FileScan parquet default.small_table1[CURNCY_ID#1263] Batched: true, DataFilters: [isnotnull(CURNCY_ID#1263)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table1], PartitionFilters: [], PushedFilters: [IsNotNull(CURNCY_ID)], ReadSchema: struct<CURNCY_ID:decimal(9,0)>, SelectedBucketsCount: 1 out of 1 : : : : : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(9,0), true] as decimal(20,0)))), [id=#297] : : : : : : : : +- *(2) Project [CURNCY_ID#1263 AS gen_attr_318#379] : : : : : : : : +- *(2) Filter isnotnull(CURNCY_ID#1263) : : : : : : : : +- *(2) ColumnarToRow : : : : : : : : +- FileScan parquet default.small_table1[CURNCY_ID#1263] Batched: true, DataFilters: [isnotnull(CURNCY_ID#1263)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table1], PartitionFilters: [], PushedFilters: [IsNotNull(CURNCY_ID)], ReadSchema: struct<CURNCY_ID:decimal(9,0)>, SelectedBucketsCount: 1 out of 1 : : : : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(4,0), true] as decimal(20,0)))), [id=#306] : : : : : : : +- *(3) Project [cntry_id#1269 AS gen_attr_317#453, rev_rollup_id#1278 AS gen_attr_315#462] : : : : : : : +- *(3) Filter isnotnull(cntry_id#1269) : : : : : : : +- *(3) ColumnarToRow : : : : : : : +- FileScan parquet default.small_table2[cntry_id#1269,rev_rollup_id#1278] Batched: true, DataFilters: [isnotnull(cntry_id#1269)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table2], PartitionFilters: [], PushedFilters: [IsNotNull(cntry_id)], ReadSchema: struct<cntry_id:decimal(4,0),rev_rollup_id:smallint> : : : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(cast(input[0, smallint, true] as int) as bigint))), [id=#315] : : : : : : +- *(4) Project [rev_rollup_id#1286 AS gen_attr_316#562, curncy_id#1289 AS gen_attr_313#565] : : : : : : +- *(4) Filter isnotnull(rev_rollup_id#1286) : : : : : : +- *(4) ColumnarToRow : : : : : : +- FileScan parquet default.small_table3[rev_rollup_id#1286,curncy_id#1289] Batched: true, DataFilters: [isnotnull(rev_rollup_id#1286)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table3], PartitionFilters: [], PushedFilters: [IsNotNull(rev_rollup_id)], ReadSchema: struct<rev_rollup_id:smallint,curncy_id:decimal(4,0)> : : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(9,0), true] as decimal(20,0)))), [id=#324] : : : : : +- *(5) Project [CURNCY_ID#1263 AS gen_attr_314#591] : : : : : +- *(5) Filter isnotnull(CURNCY_ID#1263) : : : : : +- *(5) ColumnarToRow : : : : : +- FileScan parquet default.small_table1[CURNCY_ID#1263] Batched: true, DataFilters: [isnotnull(CURNCY_ID#1263)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table1], PartitionFilters: [], PushedFilters: [IsNotNull(CURNCY_ID)], ReadSchema: struct<CURNCY_ID:decimal(9,0)>, SelectedBucketsCount: 1 out of 1 : : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(4,0), true] as decimal(20,0)))), [id=#333] : : : : +- *(6) Project [cntry_id#1269 AS gen_attr_312#665, rev_rollup_id#1278 AS gen_attr_310#674] : : : : +- *(6) Filter isnotnull(cntry_id#1269) : : : : +- *(6) ColumnarToRow : : : : +- FileScan parquet default.small_table2[cntry_id#1269,rev_rollup_id#1278] Batched: true, DataFilters: [isnotnull(cntry_id#1269)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table2], PartitionFilters: [], PushedFilters: [IsNotNull(cntry_id)], ReadSchema: struct<cntry_id:decimal(4,0),rev_rollup_id:smallint> : : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(cast(input[0, smallint, true] as int) as bigint))), [id=#342] : : : +- *(7) Project [rev_rollup_id#1286 AS gen_attr_311#774, curncy_id#1289 AS gen_attr_308#777] : : : +- *(7) Filter isnotnull(rev_rollup_id#1286) : : : +- *(7) ColumnarToRow : : : +- FileScan parquet default.small_table3[rev_rollup_id#1286,curncy_id#1289] Batched: true, DataFilters: [isnotnull(rev_rollup_id#1286)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table3], PartitionFilters: [], PushedFilters: [IsNotNull(rev_rollup_id)], ReadSchema: struct<rev_rollup_id:smallint,curncy_id:decimal(4,0)> : : +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, decimal(9,0), true] as decimal(20,0)))), [id=#351] : : +- *(8) Project [CURNCY_ID#1263 AS gen_attr_309#803] : : +- *(8) Filter isnotnull(CURNCY_ID#1263) : : +- *(8) ColumnarToRow : : +- FileScan parquet default.small_table1[CURNCY_ID#1263] Batched: true, DataFilters: [isnotnull(CURNCY_ID#1263)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table1], PartitionFilters: [], PushedFilters: [IsNotNull(CURNCY_ID)], ReadSchema: struct<CURNCY_ID:decimal(9,0)>, SelectedBucketsCount: 1 out of 1 : +- BroadcastExchange HashedRelationBroadcastMode(List(input[0, decimal(4,0), true])), [id=#360] : +- *(9) Project [cntry_id#1269, rev_rollup#1279] : +- *(9) Filter isnotnull(cntry_id#1269) : +- *(9) ColumnarToRow : +- FileScan parquet default.small_table2[cntry_id#1269,rev_rollup#1279] Batched: true, DataFilters: [isnotnull(cntry_id#1269)], Format: Parquet, Location: InMemoryFileIndex[file:/user/hive/warehouse/small_table2], PartitionFilters: [], PushedFilters: [IsNotNull(cntry_id)], ReadSchema: struct<cntry_id:decimal(4,0),rev_rollup:string> +- ReusedExchange [cntry_id#1309, rev_rollup#1319], BroadcastExchange HashedRelationBroadcastMode(List(input[0, decimal(4,0), true])), [id=#360] ``` This PR try to improve `ResolveTables` and `ResolveRelations` performance by reducing the connection times to Hive Metastore Server in such case. ### Why are the changes needed? 1. Reduce the connection times to Hive Metastore Server. 2. Improve `ResolveTables` and `ResolveRelations` performance. ### Does this PR introduce any user-facing change? No. ### How was this patch tested? manual test. After [SPARK-29606](https://issues.apache.org/jira/browse/SPARK-29606) and before this PR: ``` === Metrics of Analyzer/Optimizer Rules === Total number of runs: 9323 Total time: 2.687441263 seconds Rule Effective Time / Total Time Effective Runs / Total Runs org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveRelations 929173767 / 930133504 2 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveTables 0 / 383363402 0 / 18 org.apache.spark.sql.catalyst.optimizer.EliminateOuterJoin 0 / 99433540 0 / 4 org.apache.spark.sql.catalyst.analysis.DecimalPrecision 41809394 / 83727901 2 / 18 org.apache.spark.sql.execution.datasources.PruneFileSourcePartitions 71372977 / 71372977 1 / 1 org.apache.spark.sql.catalyst.analysis.TypeCoercion$ImplicitTypeCasts 0 / 59071933 0 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveReferences 37858325 / 58471776 5 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$PromoteStrings 20889892 / 53229016 1 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$FunctionArgumentConversion 23428968 / 50890815 1 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$InConversion 23230666 / 49182607 1 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ExtractGenerator 0 / 43638350 0 / 18 org.apache.spark.sql.catalyst.optimizer.ColumnPruning 17194844 / 42530885 1 / 6 ``` After [SPARK-29606](https://issues.apache.org/jira/browse/SPARK-29606) and after this PR: ``` === Metrics of Analyzer/Optimizer Rules === Total number of runs: 9323 Total time: 2.163765869 seconds Rule Effective Time / Total Time Effective Runs / Total Runs org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveRelations 658905353 / 659829383 2 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveTables 0 / 220708715 0 / 18 org.apache.spark.sql.catalyst.optimizer.EliminateOuterJoin 0 / 99606816 0 / 4 org.apache.spark.sql.catalyst.analysis.DecimalPrecision 39616060 / 78215752 2 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveReferences 36706549 / 54917789 5 / 18 org.apache.spark.sql.execution.datasources.PruneFileSourcePartitions 53561921 / 53561921 1 / 1 org.apache.spark.sql.catalyst.analysis.TypeCoercion$ImplicitTypeCasts 0 / 52329678 0 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$PromoteStrings 20945755 / 49695998 1 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$FunctionArgumentConversion 20872241 / 46740145 1 / 18 org.apache.spark.sql.catalyst.analysis.TypeCoercion$InConversion 19780298 / 44327227 1 / 18 org.apache.spark.sql.catalyst.analysis.Analyzer$ExtractGenerator 0 / 42312023 0 / 18 org.apache.spark.sql.catalyst.optimizer.ColumnPruning 17197393 / 39501424 1 / 6 ``` Closes #26589 from wangyum/SPARK-29947. Authored-by: Yuming Wang <yumwang@ebay.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
arjunshroff
pushed a commit
to arjunshroff/spark
that referenced
this pull request
Nov 24, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There seems to be 2 issues.
One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements.
Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED.