Question

Version Info: 
   "org.apache.storm" % "storm-core" % "1.2.1" 
   "org.apache.storm" % "storm-kafka-client" % "1.2.1"

我有3个螺栓（A，B，C）的风暴拓扑，中间螺栓平均时间约为450毫秒，其他两个螺栓不到1毫秒。

我可以使用以下并行度提示值运行拓扑：

A: 4 
B: 700
C: 10

但是当我将B的并行度提示增加到1200时，拓扑结构无法启动。

在拓扑日志中，我看到要加载执行程序的日志：B多次，如下所示：

2018-05-18 18:56:37.462 o.a.s.d.executor main [INFO] Loading executor B:[111 111]
2018-05-18 18:56:37.463 o.a.s.d.executor main [INFO] Loaded executor tasks B:[111 111]
2018-05-18 18:56:37.465 o.a.s.d.executor main [INFO] Finished loading executor B:[111 111]
2018-05-18 18:56:37.528 o.a.s.d.executor main [INFO] Loading executor B:[355 355]
2018-05-18 18:56:37.529 o.a.s.d.executor main [INFO] Loaded executor tasks B:[355 355]
2018-05-18 18:56:37.530 o.a.s.d.executor main [INFO] Finished loading executor B:[355 355]
2018-05-18 18:56:37.666 o.a.s.d.executor main [INFO] Loading executor B:[993 993]
2018-05-18 18:56:37.667 o.a.s.d.executor main [INFO] Loaded executor tasks B:[993 993]
2018-05-18 18:56:37.669 o.a.s.d.executor main [INFO] Finished loading executor B:[993 993]
2018-05-18 18:56:37.713 o.a.s.d.executor main [INFO] Loading executor B:[765 765]
2018-05-18 18:56:37.714 o.a.s.d.executor main [INFO] Loaded executor tasks B:[765 765]

但是在工作进程之间重新启动。我没有看到拓扑日志或风暴日志中的任何错误。以下是风暴日志，当工作人员重新启动时：

2018-05-18 18:51:46.755 o.a.s.d.s.Container SLOT_6700 [INFO] Killing eaf4d8ce-e758-4912-a15d-6dab8cda96d0:766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.204 o.a.s.d.s.BasicContainer Thread-7 [INFO] Worker Process 766258fe-a604-4385-8eeb-e85cad38b674 exited with code: 143
2018-05-18 18:51:47.766 o.a.s.d.s.Slot SLOT_6700 [INFO] STATE RUNNING msInState: 109081 topo:myTopology-1-1526649581 worker:766258fe-a604-4385-8eeb-e85cad38b674 -> KILL msInState: 0 topo:myTopology-1-1526649581 worker:766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.766 o.a.s.d.s.Container SLOT_6700 [INFO] GET worker-user for 766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.774 o.a.s.d.s.Slot SLOT_6700 [WARN] SLOT 6700 all processes are dead...
2018-05-18 18:51:47.775 o.a.s.d.s.Container SLOT_6700 [INFO] Cleaning up eaf4d8ce-e758-4912-a15d-6dab8cda96d0:766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.775 o.a.s.d.s.Container SLOT_6700 [INFO] GET worker-user for 766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.775 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers/766258fe-a604-4385-8eeb-e85cad38b674/pids/27798
2018-05-18 18:51:47.775 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers/766258fe-a604-4385-8eeb-e85cad38b674/heartbeats
2018-05-18 18:51:47.780 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers/766258fe-a604-4385-8eeb-e85cad38b674/pids
2018-05-18 18:51:47.780 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers/766258fe-a604-4385-8eeb-e85cad38b674/tmp
2018-05-18 18:51:47.781 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers/766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.782 o.a.s.d.s.Container SLOT_6700 [INFO] REMOVE worker-user 766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.782 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/workers-users/766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.783 o.a.s.d.s.BasicContainer SLOT_6700 [INFO] Removed Worker ID 766258fe-a604-4385-8eeb-e85cad38b674
2018-05-18 18:51:47.783 o.a.s.l.AsyncLocalizer SLOT_6700 [INFO] Released blob reference myTopology-1-1526649581 6700 Cleaning up BLOB references...
2018-05-18 18:51:47.784 o.a.s.l.AsyncLocalizer SLOT_6700 [INFO] Released blob reference myTopology-1-1526649581 6700 Cleaning up basic files...
2018-05-18 18:51:47.785 o.a.s.d.s.AdvancedFSOps SLOT_6700 [INFO] Deleting path /home/saurabh/storm-run/supervisor/stormdist/myTopology-1-1526649581
2018-05-18 18:51:47.808 o.a.s.d.s.Slot SLOT_6700 [INFO] STATE KILL msInState: 42 topo:myTopology-1-1526649581 worker:null -> EMPTY msInState: 0

这种情况一直在发生，拓扑结构永远不会重新开始，当并行性提示为bolt时，它常常完美地启动：B为700，没有其他变化。

我在这里看到一个有趣的日志，还不确定这意味着什么：

工作流程766258fe-a604-4385-8eeb-e85cad38b674退出，代码：143

任何建议？

修改

配置：

topology.worker.childopts: -Xms1g -Xmx16g
topology.worker.logwriter.childopts: -Xmx1024m
topology.worker.max.heap.size.mb: 3072.0
worker.childopts: -Xms1g -Xmx16g -Dcom.sun.management.jmxremote -Dcom.sun.management.jmxremote.port=1%ID% -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.ssl=false -XX:+UseG1GC -XX:+AggressiveOpts -XX:+PrintGCDetails -XX:+PrintGCDateStamps -Xloggc:/home/saurabh.mimani/apache-storm-1.2.1/logs/gc.log -XX:+UseGCLogFileRotation -XX:NumberOfGCLogFiles=10 -XX:GCLogFileSize=1M -Dorg.newsclub.net.unix.library.path=/usr/share/specter/uds-lib/
worker.gc.childopts:
worker.heap.memory.mb: 8192
supervisor.childopts: -Xms1g -Xmx16g

修改

在gist中记录strace -fp PID -e trace=read,write,network,signal,ipc。

还没有完全理解，有些相关的看法：

[pid 3362] open（＆＃34; / usr / lib / locale / UTF-8 / LC_CTYPE＆＃34;，O_RDONLY）= -1 ENOENT（没有这样的文件或目录）

[pid 3362] kill（1487，SIGTERM）= 0

[pid 3362] close（1）

Answer 1

快速谷歌建议143是JVM收到SIGTERM时的退出代码（例如Always app Java end with "Exit 143" Ubuntu）。您可能正在耗尽内存，或者操作系统可能因某些其他原因而终止该进程。请记住，将并行度提示设置为1200意味着您将获得螺栓B的1200个任务（副本），之前您只有700个任务。

Answer 2

我能够通过调整以下配置来实现这一点，似乎是由于nimbus.task.launch.sec导致超时，设置为120并且如果它在120秒内未启动则重启工作者。 / p>

更新了其中一些配置的值：

drpc.request.timeout.secs: 1600
supervisor.worker.start.timeout.secs: 1200
nimbus.supervisor.timeout.secs: 1200
nimbus.task.launch.secs: 1200

关于nimbus.task.launch.sec：

最初启动任务时使用的特殊超时。在启动期间，这是在第一次心跳之前使用的超时，覆盖nimbus.task.timeout.secs。启动时存在单独的超时，因为启动新JVM并配置它们可能会有相当多的开销。

风暴拓扑并不是以1200的并行性暗示开始的

2 个答案: