sparksql 复杂查询OOM
Posted Books&Coffee
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了sparksql 复杂查询OOM相关的知识，希望对你有一定的参考价值。
平台上执行复杂查询，OOM，根据日志提示的结局方法：
-- SET spark.driver.memory=6/8G;【还是OOM】
set spark.sql.autoBroadcastJoinThreshold=-1;【解决问题】
Exception in thread "broadcast-exchange-0" java.lang.OutOfMemoryError: Not enough memory to build and broadcast the table to all worker nodes. As a workaround, you can either disable broadcast by setting spark.sql.autoBroadcastJoinThreshold to -1 or increase the spark driver memory by setting spark.driver.memory to a higher value at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1$$anonfun$apply$1.apply(BroadcastExchangeExec.scala:115) at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1$$anonfun$apply$1.apply(BroadcastExchangeExec.scala:73) at org.apache.spark.sql.execution.SQLExecution$.withExecutionId(SQLExecution.scala:97) at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1.apply(BroadcastExchangeExec.scala:72) at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1.apply(BroadcastExchangeExec.scala:72) at scala.concurrent.impl.Future$PromiseCompletingRunnable.liftedTree1$1(Future.scala:24) at scala.concurrent.impl.Future$PromiseCompletingRunnable.run(Future.scala:24) at java.lang.Thread.run(Thread.java:748)
2019-11-20 13:45:38 调度器收到作业实例: T_6466665892523663_20191120134538_2, 加入到时间调度队列中.
2019-11-20 13:45:38 计划开始时间是: 2019-11-19 00:10:00, 等待调度中...
2019-11-20 13:45:38 实例到达调度时间, 但父节点尚未完成
2019-11-20 13:47:05 实例的父节点已完成, 且达调度时间，加入调度队列.
2019-11-20 13:47:05 当前有0个作业排队中...
2019-11-20 13:47:05 等待调度中...
2019-11-20 13:47:05 开始调度作业实例
2019-11-20 13:47:05 第1次在节点: 10.94.0.15 上获取Yarn资源信息.
2019-11-20 13:47:05 Yarn:http://10.94.0.16:8088,10.94.0.10:8088, 队列:root.hive 的剩余资源是:"VCores: 17, Memory: 22796M".
2019-11-20 13:47:05 获取Yarn资源结束,开始下放到Node上执行.
2019-11-20 13:47:05 第1次调度到节点: 10.94.0.15 成功.
the current free memorySize is 7240m, greater than the threshold 64m, satisfying the memory requirement to execute task
获取任务插件: SparkSql插件 成功。插件信息：
引擎：SparkSql
引擎适配版本：[]
厂商：
厂商版本：
start to init task.
获取到 resource-plugin: sparkSql2.2.0-cdh5.12.0-1.0.0-jar-with-dependencies.jar
19/11/20 13:47:06 INFO spark.SparkContext: Running Spark version 2.3.0.cloudera2
19/11/20 13:47:06 INFO spark.SparkContext: Submitted application: dwd_mem_2b_applet_filter_task_detail_walker_T_6466665892523663_20191120134538_2_1
19/11/20 13:47:07 INFO server.Server: jetty-9.3.z-SNAPSHOT
19/11/20 13:47:08 INFO client.ConfiguredRMFailoverProxyProvider: Failing over to rm177
19/11/20 13:47:08 INFO yarn.Client: Requesting a new application from cluster with 5 NodeManagers
19/11/20 13:47:08 INFO yarn.Client: Verifying our application has not requested more than the maximum memory capability of the cluster (8192 MB per container)
19/11/20 13:47:08 INFO yarn.Client: Will allocate AM container, with 896 MB memory including 384 MB overhead
19/11/20 13:47:08 INFO yarn.Client: Setting up container launch context for our AM
19/11/20 13:47:08 INFO yarn.Client: Setting up the launch environment for our AM container
19/11/20 13:47:08 INFO yarn.Client: Preparing resources for our AM container
19/11/20 13:47:09 INFO yarn.Client: Submitting application application_1573451947838_0670 to ResourceManager
19/11/20 13:47:09 INFO impl.YarnClientImpl: Submitted application application_1573451947838_0670
19/11/20 13:47:10 INFO yarn.Client: Application report for application_1573451947838_0670 (state: ACCEPTED)
19/11/20 13:47:10 INFO yarn.Client:
client token: N/A
diagnostics: N/A
ApplicationMaster host: N/A
ApplicationMaster RPC port: -1
queue: root.hive
start time: 1574228829689
final status: UNDEFINED
tracking URL: http://cdh-hadoop-16:8088/proxy/application_1573451947838_0670/
user: deploy
19/11/20 13:47:11 INFO yarn.Client: Application report for application_1573451947838_0670 (state: ACCEPTED)
19/11/20 13:47:12 INFO yarn.Client: Application report for application_1573451947838_0670 (state: ACCEPTED)
19/11/20 13:47:13 INFO yarn.Client: Application report for application_1573451947838_0670 (state: RUNNING)
19/11/20 13:47:13 INFO yarn.Client:
client token: N/A
diagnostics: N/A
ApplicationMaster host: 10.94.0.16
ApplicationMaster RPC port: 0
queue: root.hive
start time: 1574228829689
final status: UNDEFINED
tracking URL: http://cdh-hadoop-16:8088/proxy/application_1573451947838_0670/
user: deploy
19/11/20 13:47:13 INFO storage.BlockManager: Using org.apache.spark.storage.RandomBlockReplicationPolicy for block replication policy
19/11/20 13:47:13 INFO storage.BlockManager: external shuffle service port = 7337
19/11/20 13:47:13 INFO storage.BlockManager: Initialized BlockManager: BlockManagerId(driver, shuqi-node, 36374, None)
19/11/20 13:47:14 INFO scheduler.EventLoggingListener: Logging events to hdfs://hahdfs/user/spark/spark2ApplicationHistory/application_1573451947838_0670
19/11/20 13:47:14 INFO spark.SparkContext: Registered listener com.cloudera.spark.lineage.NavigatorAppListener
19/11/20 13:47:14 INFO internal.SharedState: loading hive config file: file:/etc/hadoop/conf.cloudera.yarn/hive-site.xml
19/11/20 13:47:14 INFO internal.SharedState: spark.sql.warehouse.dir is not set, but hive.metastore.warehouse.dir is set. Setting spark.sql.warehouse.dir to the value of hive.metastore.warehouse.dir (‘/user/hive/warehouse‘).
19/11/20 13:47:14 INFO internal.SharedState: Warehouse path is ‘/user/hive/warehouse‘.
19/11/20 13:47:14 INFO state.StateStoreCoordinatorRef: Registered StateStoreCoordinator endpoint
19/11/20 13:47:15 INFO hive.metastore: Trying to connect to metastore with URI thrift://cdh-hadoop-16:9083
19/11/20 13:47:15 INFO hive.metastore: Opened a connection to metastore, current connections: 1
19/11/20 13:47:15 INFO hive.metastore: Connected to metastore.
19/11/20 13:47:15 WARN command.SetCommand: ‘SET hive.exec.dynamic.partition.mode=nonstrict‘ might not work, since Spark doesn‘t support changing the Hive config dynamically. Please pass the Hive-specific config by adding the prefix spark.hadoop (e.g. spark.hadoop.hive.exec.dynamic.partition.mode) when starting a Spark application. For details, see the link: https://spark.apache.org/docs/latest/configuration.html#dynamically-loading-spark-properties.
开始执行:
insert overwrite table dwd_mem_2b_applet_filter_task_detail
select t1.ID as member_id,
t2.LeaguerID as leaguer_id,
t9.member_employee_id as member_employee_id,
t2.ShopID as member_store_id,
t1.Avatar as member_avatar,
t1.RealName as member_name,
t1.Phone as member_phone,
t1.Birthday as member_birthday,
t12.outbound_employee_id as outbound_employee_id,
t12.outbound_member_phone as outbound_member_phone,
t12.outbound_time as outbound_time,
t4.baby_id as baby_id,
t4.baby_avatar as baby_avatar,
t4.baby_name as baby_name,
t4.baby_gender as baby_gender,
t4.baby_age as baby_age,
t4.baby_birthday as baby_birthday,
t1.register_2_nowadays as register_2_nowadays,
t10.register_channel as register_channel,
case when t3.first_card_open_date is not null then 1
else 0 end as is_open_cards,
t3.first_card_open_date as first_card_open_date,
t2.LeaguerCode as card_id,
t15.consume_frequency_dateCard as consume_frequency_dateCard,
t16.consume_frequency_no_dateCard as consume_frequency_no_dateCard,
t5.card_expire_date as card_expire_date,
t6.total_recharge as total_recharge,
t7.total_consume_coin as total_consume_coin,
t8.last_consume_date as last_consume_date,
t8.last_consume_2_nowadays as last_consume_2_nowadays,
t5.card_times_left as card_times_left,
t2.card_coin_left as card_coin_left,
t5.card_days_left as card_days_left,
t11.card_coupon_days_left as card_coupon_days_left,
t6.last_recharge_2_nowadays as last_recharge_2_nowadays,
t2.card_coin_left/t13.card_coin_left as left_coin_ratio,
t5.card_times_left/t14.card_times_left as left_times_ratio
from
(
select ID,
Avatar,
RealName,
Phone,
substr(to_date(Birthday),-5,5) as Birthday,
datediff(CURRENT_DATE,to_date(CreateTime)) as register_2_nowadays -- 注册距今天数
from ods_cuihua_dqmly_dy_sp_leaguerBase
where ds= ‘20191119‘
) t1
left join
(
select p1.ID as LeaguerID, -- 门店会员编码(系统自动生成 ID)
p1.LeaguerBaseId, -- 会员编码
p1.ShopID,
p3.LeaguerCode, -- 会员卡号
p2.card_coin_left -- 币剩余数量
from ods_cuihua_dqmly_dy_sp_Leaguer p1
left join
(
select ID as LeaguerId_p2,
nvl(SUM(CoinBal1) + SUM(GiveCoinBal1)+ SUM(GiveBCoinBal1),0) as card_coin_left -- 币剩余数量
from ods_cuihua_dqmly_dy_sp_Leaguer
where ds= ‘20191119‘
group by ID
) p2
on p1.ID = p2.LeaguerId_p2
left join
(
select LeaguerId as LeaguerId_p3,
LeaguerCode
from ods_cuihua_dqmly_dy_sp_LeaguerCard
where ds= ‘20191119‘
) p3
on p1.ID = p3.LeaguerId_p3
where p1.ds= ‘20191119‘
) t2
on t1.ID = t2.LeaguerBaseId
left join
(
select p1.LeaguerBaseId as LeaguerBaseId_t3,
min(p2.CreateTime) as first_card_open_date -- 首次开卡时间
from ods_cuihua_dqmly_dy_sp_Leaguer p1
left join
(
select LeaguerId,
LeaguerCode,
CreateTime
from ods_cuihua_dqmly_dy_sp_LeaguerCard
where ds= ‘20191119‘
) p2
on p1.ID = p2.LeaguerId
where p1.ds= ‘20191119‘
group by p1.LeaguerBaseId
) t3
on t1.ID = t3.LeaguerBaseId_t3
left join
(
select ID as baby_id,
LeaguerBaseId as LeaguerBaseId_t4,
Avatar as baby_avatar,
RealName as baby_name,
Sex as baby_gender,
cast(datediff(CURRENT_DATE,to_date(Birthday))/365 as decimal(10,1)) as baby_age,
substr(to_date(Birthday),-5,5) as baby_birthday
from ods_cuihua_dqmly_dy_sp_leaguerbaby
where ds= ‘20191119‘
and ID is not null
and LeaguerBaseId is not null
) t4
on t1.ID = t4.LeaguerBaseId_t4
left join
(
select tc.LeaguerID as LeaguerID_t5,
tc.LeaguerCode as LeaguerCode_t5,
te.card_expire_time as card_expire_date, -- 卡到期日期
nvl(tf.card_times_left,0) as card_times_left, -- 次卡剩余次数
nvl(datediff(to_date(te.card_expire_time),CURRENT_DATE),0) as card_days_left -- 计时卡剩余天数
from
(
select LeaguerId,
LeaguerCode,
ID as LeaguerCardId
from ods_cuihua_dqmly_dy_sp_LeaguerCard
where ds= ‘20191119‘
) tc
left join
(
select LeaguerID as LeaguerID_te,
LeaguerCardId as LeaguerCardId_te,
case when ReExpireTime is null then EndTime
when EndTime <= ReExpireTime then ReExpireTime
end as card_expire_time
from ods_cuihua_dqmly_dy_sp_LeaguerTimeCard
where ds= ‘20191119‘
and LeaguerCardId is not null
) te
on tc.LeaguerId = te.LeaguerID_te and tc.LeaguerCardId = te.LeaguerCardId_te
inner join
(
select LeaguerID,
SUM(EnabledNum) as card_times_left
from ods_cuihua_dqmly_dy_sp_LeaguerTimeCard
where ds= ‘20191119‘
and (CURRENT_TIMESTAMP < EndTime or CURRENT_TIMESTAMP < ReExpireTime )
and EnabledNum > 0
group by LeaguerID
) tf
on te.LeaguerID_te = tf.LeaguerID
) t5
on t2.LeaguerID = t5.LeaguerID_t5 and t2.LeaguerCode = t5.LeaguerCode_t5
left join
(
select LeaguerID as LeaguerID_t6,
sum(money/100) as total_recharge, -- 总充值金额
datediff(CURRENT_DATE,to_date(max(createtime))) as last_recharge_2_nowadays -- 最后充值时间距今天数
from ods_cuihua_dqmly_dy_sp_setMealSellLog_delta
where LeaguerID is not null
group by LeaguerID
) t6
on t2.LeaguerID = t6.LeaguerID_t6
left join
(
select p.LeaguerID as LeaguerID_t7,
nvl(sum(p.total_consume_coin),0) as total_consume_coin
from
(
select LeaguerID,
sum(coinnum) as total_consume_coin -- 总耗币数:消费币,兑换礼品,自取币,币换年票,赠币过期,币过户等
from ods_cuihua_dqmly_dy_sp_playLog_delta
group by LeaguerID
union all
select LeaguerID,
sum(SurplusCoin1) as total_consume_coin
from ods_cuihua_dqmly_dy_sp_LeaguerGiveCoinLog_delta
where IsExpire=‘true‘ and SurplusCoin1>0 -- (IsExpire:是否过期、SurplusCoin1:过期的币数)
group by LeaguerID
union all
select LeaguerID,
sum(UsedValue) as total_consume_coin
from ods_cuihua_dqmly_dy_sp_giftSellLog_delta
where UseValueType=5 -- (UseValueType=5表示币兑换礼品 UsedValue:消耗币数)
group by LeaguerID
union all
select LeaguerID,
sum(abs(CoinNumber1)) as total_consume_coin
from ods_cuihua_dqmly_dy_sp_setMealSellLog_delta
where PayType=7 -- (PayType=7 表示通过币购买套餐)
group by LeaguerID
union all
select FromLeaguer as LeaguerID,
sum(Amount) as total_consume_coin
from ods_cuihua_dqmly_dy_sp_valueTransferLog_delta -- (FromLeaguer:过户源用户 ToLeaguer:过户目标用户)
where ValueType = 1
group by FromLeaguer
union all
select LeaguerID,
sum(RealCoinNum) as total_consume_coin
from ods_cuihua_dqmly_dy_sp_getCoin -- (RealCoinNum:实际提币数量)
where ds= ‘20191119‘
group by LeaguerID
) p
where p.LeaguerID is not null
group by p.LeaguerID
) t7
on t2.LeaguerID = t7.LeaguerID_t7
left join
(
select LeaguerID as LeaguerID_t8,
max(CreateTime) as last_consume_date, -- 最后消费日期
datediff(CURRENT_DATE,to_date(max(CreateTime))) as last_consume_2_nowadays -- 最后一次消费距今天数:取 耗币、刷计时卡、耗次、核券的最后一次时间
from ods_cuihua_dqmly_dy_sp_playLog_delta
where LeaguerID is not null
group by LeaguerID
) t8
on t2.LeaguerID = t8.LeaguerID_t8
left join
(
select p1.invite_member_no as LeaguerBaseId_t9,
p2.employee_id as member_employee_id
from
(
select invite_code,
invite_member_no
from ods_mengguang_xc_crm_member_db1_member_invite_summary
where ds= ‘20191119‘
and invite_code is not null
and invite_member_no is not null
) p1
inner join
(
select invite_code as invite_code_p2,
employee_id
from ods_mengguang_xc_activity_manage_db_employee_invite_code
where ds= ‘20191119‘
) p2
on p1.invite_code = p2.invite_code_p2
) t9
on t1.ID = t9.LeaguerBaseId_t9
left join
(
select channel_info as register_channel,
member_no as LeaguerBaseId_t10
from ods_mengguang_xc_crm_member_db1_member_profile
where ds= ‘20191119‘
and member_no is not null
) t10
on t1.ID = t10.LeaguerBaseId_t10
left join
(
select p1.LeaguerBaseId as LeaguerBaseId_t11,
case when p2.draw_limit_day <= 0 then datediff(to_date(p2.end_date),CURRENT_DATE)
else datediff(date_add(to_date(p1.date_created), p2.draw_limit_day),CURRENT_DATE) end as card_coupon_days_left
from
(
select coupon_id,
member_id as LeaguerBaseId,
date_created
from ods_qunshuo_amazingland_campaign_coupon_code_event
where ds= ‘20191119‘
and is_delete = ‘NO‘
and coupon_status in(‘WAIT_RECEIVE‘,‘WAIT_USE‘)
)p1
inner join
(
select coupon_number,
end_date,
draw_limit_day
from ods_qunshuo_amazingland_campaign_coupon
where ds= ‘20191119‘
and is_delete = ‘NO‘
and coupon_number <> ‘‘
) p2
on p1.coupon_id = p2.coupon_number
) t11
on t1.ID = t11.LeaguerBaseId_t11
left join
(
select tb1.custNum as outbound_member_id, -- 外呼会员编码，uuid
tb1.staffNum as outbound_employee_id, -- 外呼员工id，不一定是会员所属员工id,有可能是别的员工外呼
tb1.custPhone as outbound_member_phone, -- 外呼会员手机号码
tb2.outbound_time
from ods_lixiang_immccdb_callout_info tb1
inner join
(
select custNum as custNum_tb2, -- 会员编码
max(createTime) as outbound_time -- 最后一次外呼时间
from ods_lixiang_immccdb_callout_info
where ds= ‘20191119‘
and call_status = 0 -- 呼叫成功 【理想插入数据之后，需要打开这个限制条件】
group by custNum
) tb2
on tb1.custNum = tb2.custNum_tb2
where tb1.ds= ‘20191119‘
) t12
on t1.ID = t12.outbound_member_id
left join
(
select LeaguerID as LeaguerID_t13,
card_coin_left
from dwd_card_coin_left_last_recharge
group by LeaguerID,card_coin_left
) t13
on t2.LeaguerID = t13.LeaguerID_t13
left join
(
select LeaguerID as LeaguerID_14,
card_times_left
from dwd_card_times_left_last_recharge
group by LeaguerID,card_times_left
) t14
on t2.LeaguerID = t14.LeaguerID_14
left join
(
SELECT ta.LeaguerID as LeaguerID_15,
count(to_date(ta.createTime)) as consume_frequency_dateCard
FROM ods_cuihua_dqmly_dy_sp_playLog_delta AS ta
INNER JOIN ods_cuihua_dqmly_dy_sp_timeCard AS tb
ON ta.TimeCardId = tb.ID
where tb.ds= ‘20191119‘
and tb.Category IN (2,3) -- 保证是计时卡会员
and to_date(ta.createtime) >= date_sub(CURRENT_DATE,180)
and to_date(ta.createtime) <= CURRENT_DATE
group by ta.LeaguerID
)t15
on t2.LeaguerID = t15.LeaguerID_15
left join
(
SELECT ta.LeaguerID as LeaguerID_16,
count(to_date(ta.createTime)) as consume_frequency_no_dateCard
FROM ods_cuihua_dqmly_dy_sp_playLog_delta AS ta
INNER JOIN ods_cuihua_dqmly_dy_sp_timeCard AS tb
ON ta.TimeCardId = tb.ID
where tb.ds= ‘20191119‘
and tb.Category NOT IN (2,3) -- 保证是非计时卡会员
and to_date(ta.createtime) >= date_sub(CURRENT_DATE,180)
and to_date(ta.createtime) <= CURRENT_DATE
group by ta.LeaguerID
)t16
on t2.LeaguerID = t16.LeaguerID_16;
19/11/20 13:47:17 INFO hive.HiveMetastoreCatalog: Inferring case-sensitive schema for table xcdqm_prd.ods_cuihua_dqmly_dy_sp_leaguerbase (inference mode: INFER_AND_SAVE)
19/11/20 13:47:17 INFO scheduler.DAGScheduler: Parents of final stage: List()
19/11/20 13:47:17 INFO scheduler.DAGScheduler: Missing parents: List()
19/11/20 13:47:17 INFO spark.SparkContext: Created broadcast 0 from broadcast at DAGScheduler.scala:1039
19/11/20 13:47:17 INFO cluster.YarnScheduler: Adding task set 0.0 with 1 tasks
19/11/20 13:47:18 INFO spark.ExecutorAllocationManager: Requesting 1 new executor because tasks are backlogged (new desired total will be 1)
Exception in thread "broadcast-exchange-0" java.lang.OutOfMemoryError: Not enough memory to build and broadcast the table to all worker nodes. As a workaround, you can either disable broadcast by setting spark.sql.autoBroadcastJoinThreshold to -1 or increase the spark driver memory by setting spark.driver.memory to a higher value
at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1$$anonfun$apply$1.apply(BroadcastExchangeExec.scala:115)
at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1$$anonfun$apply$1.apply(BroadcastExchangeExec.scala:73)
at org.apache.spark.sql.execution.SQLExecution$.withExecutionId(SQLExecution.scala:97)
at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1.apply(BroadcastExchangeExec.scala:72)
at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec$$anonfun$relationFuture$1.apply(BroadcastExchangeExec.scala:72)
at scala.concurrent.impl.Future$PromiseCompletingRunnable.liftedTree1$1(Future.scala:24)
at scala.concurrent.impl.Future$PromiseCompletingRunnable.run(Future.scala:24)
at java.lang.Thread.run(Thread.java:748)
以上是关于sparksql 复杂查询OOM的主要内容，如果未能解决你的问题，请参考以下文章