Cassandra 压力 "Failed to connect over JMX; not collecting these stats"

Question

我今天第一次尝试使用 cassandra-stress 工具。尽管我能够运行该工具，但输出中会显示很多“无法通过 JMX 连接；未收集这些统计信息”消息

命令

cassandra-stress user \
    profile=./stress_write.yaml ops\(insert=1\) \
    n=1000000 \
    -log file=./stress_write.log \
    -node node1,node2,node3,node4,node5,node6

输出

WARN  19:44:25 Found host with 0.0.0.0 as rpc_address, using listen_address (/node5) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:25 Found host with 0.0.0.0 as rpc_address, using listen_address (/node1) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:25 Found host with 0.0.0.0 as rpc_address, using listen_address (/node2) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:25 Found host with 0.0.0.0 as rpc_address, using listen_address (/node4) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:25 Found host with 0.0.0.0 as rpc_address, using listen_address (/node3) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:26 Found host with 0.0.0.0 as rpc_address, using listen_address (/node5) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:26 Found host with 0.0.0.0 as rpc_address, using listen_address (/node1) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:26 Found host with 0.0.0.0 as rpc_address, using listen_address (/node2) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:26 Found host with 0.0.0.0 as rpc_address, using listen_address (/node4) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
WARN  19:44:26 Found host with 0.0.0.0 as rpc_address, using listen_address (/node3) to contact it instead. If this is incorrect you should avoid the use of 0.0.0.0 server side.
INFO  19:44:26 Using data-center name 'DC2' for DCAwareRoundRobinPolicy (if this is incorrect, please provide the correct datacenter name with DCAwareRoundRobinPolicy constructor)
INFO  19:44:26 New Cassandra host /node2:9042 added
INFO  19:44:26 New Cassandra host /node5:9042 added
Connected to cluster: MyCluster
INFO  19:44:26 New Cassandra host /node4:9042 added
INFO  19:44:26 New Cassandra host /node1:9042 added
INFO  19:44:26 New Cassandra host /node6:9042 added
Datatacenter: DC2; Host: /node4; Rack: rack1
Datatacenter: DC2; Host: /node3; Rack: rack1
Datatacenter: DC2; Host: /node6; Rack: rack1
Datatacenter: DC2; Host: /node5; Rack: rack1
Datatacenter: DC2; Host: /node1; Rack: rack1
Datatacenter: DC2; Host: /node2; Rack: rack1
INFO  19:44:26 New Cassandra host /node3:9042 added
Created schema. Sleeping 6s for propagation.
Failed to connect over JMX; not collecting these stats
Generating batches with [1..1] partitions and [1..1] rows (of [1..1] total rows in the partitions)
Failed to connect over JMX; not collecting these stats
Failed to connect over JMX; not collecting these stats
Improvement over 4 threadCount: 36%
Failed to connect over JMX; not collecting these stats
Improvement over 8 threadCount: 138%
Failed to connect over JMX; not collecting these stats
Improvement over 16 threadCount: 48%
Failed to connect over JMX; not collecting these stats
Improvement over 24 threadCount: 33%
Failed to connect over JMX; not collecting these stats
Improvement over 36 threadCount: 27%
Failed to connect over JMX; not collecting these stats
Improvement over 54 threadCount: 39%
Failed to connect over JMX; not collecting these stats
Improvement over 81 threadCount: 37%
Failed to connect over JMX; not collecting these stats
Improvement over 121 threadCount: 16%
Failed to connect over JMX; not collecting these stats
Improvement over 181 threadCount: 1%
Failed to connect over JMX; not collecting these stats
Improvement over 271 threadCount: 15%
Failed to connect over JMX; not collecting these stats
Improvement over 406 threadCount: 3%
Failed to connect over JMX; not collecting these stats
Improvement over 609 threadCount: -3%

是否需要为 JMX 指定任何命令行或基于文件的配置参数？我已经测试并确认压力机和我的节点之间的连接不是问题，因为我能够通过 jmxsh 在它们之间建立连接。

输出的另一个问题(可能与 JMX 错误有关，也可能无关)是它缺少一些关键部分。我引用了这个 Datastax documentation page 的示例输出显示我得到的东西中缺少的部分:

WARNING: uncertainty mode (err<) results in uneven workload between thread runs, so should be used for high level analysis only
Running with 4 threadCount
Running WRITE with 4 threads until stderr of mean < 0.02
total ops , adj row/s,    op/s,    pk/s,   row/s,    mean,     med,     .95,     .99,    .999,     max,   time,   stderr,  gc: #,  max ms,  sum ms,  sdv ms,      mb
2552      ,      2553,    2553,    2553,    2553,     1.5,     1.4,     2.5,     6.0,    12.6,    18.0,    1.0,  0.00000,      0,       0,       0,       0,       0
5173      ,      2634,    2613,    2613,    2613,     1.5,     1.5,     1.8,     2.6,     8.6,     9.2,    2.0,  0.00000,      0,       0,       0,       0,       0
...

Results:
op rate                   : 3954
partition rate            : 3954
row rate                  : 3954
latency mean              : 1.0
latency median            : 0.8
latency 95th percentile   : 1.5
latency 99th percentile   : 1.8
latency 99.9th percentile : 2.2
latency max               : 73.6
total gc count            : 25
total gc mb               : 1826
total gc time (s)         : 1
avg gc time(ms)           : 37
stdev gc time(ms)         : 10
Total operation time      : 00:00:59
Sleeping for 15s
Running with 4 threadCount

备注

我的集群正在运行 DSE 4.6.1 (Cassandra 2.0.12)

我正在另一台机器上运行压力工具

压力工具版本来自 DS C 2.1( Cassandra 2.1)

Answer 1

我有相同的设置(Cassandra 版本是 2.0.12)并且压力工具来自 2.1 并且看到了类似的问题。
终于有时间去调查了。

我下载了源代码并在调试器中运行它。我看到的是此错误消息具有误导性。该工具连接到 JMX，但其中一个 mBean ( org.apache.cassandra.service:type=GCInspector ) 有问题。

当我使用以下选项运行压力测试时，我看到了同样的异常:-log level=verbose并看到以下异常:

java.lang.reflect.UndeclaredThrowableException
        at com.sun.proxy.$Proxy11.getAndResetStats(Unknown Source)
        at org.apache.cassandra.tools.NodeProbe.getAndResetGCStats(NodeProbe.java:385)
        at org.apache.cassandra.stress.util.JmxCollector.<init>(JmxCollector.java:86)
        at org.apache.cassandra.stress.StressMetrics.<init>(StressMetrics.java:64)
        at org.apache.cassandra.stress.StressAction.run(StressAction.java:187)
        at org.apache.cassandra.stress.StressAction.warmup(StressAction.java:97)
        at org.apache.cassandra.stress.StressAction.run(StressAction.java:61)
        at org.apache.cassandra.stress.Stress.main(Stress.java:109)
Caused by: javax.management.InstanceNotFoundException: org.apache.cassandra.service:type=GCInspector 
        at com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.getMBean(Unknown Source)
        at ....

我使用 jConsole 连接到 Cassandra，版本 2.0.12 没有这个 mBean。

但是我的输出包含样本中引用的大部分数据(垃圾收集统计数据除外)。

您是否尝试过使用默认配置运行 cassandra-stress？还可以尝试为日志设置详细信息，可能会给你一些想法。