extract workers' resources stats

[Stefanos Antaris]

Hi to all.

I am working an a research project using flink and i would like to extract the CPU + RAM resources consumed on each worker in order to include the stats in my paper. Can anyone  advice me on how could i extract them?

Thanks in advance.
Stefanos

[Andra Lungu]

Hi,

You can use something like iostat to extract the CPU usage.

For instance, I call this script on the JM node:
#!/usr/bin/env bash

lines=`cat /home/andra.lungu/hostnames.txt | paste -d, -s`
echo "${lines}"
IFS="," read -ra hostnames <<<"${lines}"
for line in "${hostnames[@]}"; do
  ssh ${line} `iostat -xtc 5 3 > /home/andra.lungu/cpu_stats/${line}` &
done

And then I parse the individual output in the cpu_stats folder with python to form charts.

To my knowledge. Flink does not do something like that for you, at least not yet :)

Bottom line is that you can use linux commands: iostat, dstat, htop, etc.

Hope this helps!

Andra

On Fri, Aug 7, 2015 at 5:07 PM, Stefanos Antaris <[hidden email]> wrote:

Hi to all.

I am working an a research project using flink and i would like to extract the CPU + RAM resources consumed on each worker in order to include the stats in my paper. Can anyone  advice me on how could i extract them?

Thanks in advance.
Stefanos

[rmetzger0]

Hi Stefanos,

you can also write yourself a little script/tool which is periodically requesting the following JSON from the JobManager:

http://localhost:8081/setupInfo?get=taskmanagers&_=1438972693441

It returns a JSON string like this:

{"taskmanagers":[{"path":"akka:\/\/flink\/user\/taskmanager","dataPort":51695,"timeSinceLastHeartbeat":3,"slotsNumber":1,"freeSlots":1,"cpuCores":8,"physicalMemory":15941,"freeMemory":736,"managedMemory":461,"instanceID":"187cdbc1de0acdb02b5bdc796cefceaf","metrics":{"version":"3.0.0","gauges":{"cpuLoad":{"value":0.09425878320479862},"gc.PS-MarkSweep.count":{"value":6},"gc.PS-MarkSweep.time":{"value":764},"gc.PS-Scavenge.count":{"value":3},"gc.PS-Scavenge.time":{"value":121},"load":{"value":3.04},"memory.heap.committed":{"value":771751936},"memory.heap.init":{"value":805306368},"memory.heap.max":{"value":771751936},"memory.heap.usage":{"value":0.741875586302384},"memory.heap.used":{"value":572543920},"memory.non-heap.committed":{"value":50790400},"memory.non-heap.init":{"value":24576000},"memory.non-heap.max":{"value":318767104},"memory.non-heap.usage":{"value":0.09098903756392629},"memory.non-heap.used":{"value":29004312},"memory.pools.Code-Cache.usage":{"value":0.020928700764973957},"memory.pools.PS-Eden-Space.usage":{"value":0.1786344051361084},"memory.pools.PS-Old-Gen.usage":{"value":0.9994582533836365},"memory.pools.PS-Perm-Gen.usage":{"value":0.10412535071372986},"memory.pools.PS-Survivor-Space.usage":{"value":0},"memory.total.committed":{"value":822542336},"memory.total.init":{"value":829882368},"memory.total.max":{"value":1090519040},"memory.total.used":{"value":601962568}},"counters":{},"histograms":{},"meters":{},"timers":{}}}]}

On Fri, Aug 7, 2015 at 5:43 PM, Andra Lungu <[hidden email]> wrote:

Hi,

You can use something like iostat to extract the CPU usage.

For instance, I call this script on the JM node:
#!/usr/bin/env bash

lines=`cat /home/andra.lungu/hostnames.txt | paste -d, -s`
echo "${lines}"
IFS="," read -ra hostnames <<<"${lines}"
for line in "${hostnames[@]}"; do
  ssh ${line} `iostat -xtc 5 3 > /home/andra.lungu/cpu_stats/${line}` &
done

And then I parse the individual output in the cpu_stats folder with python to form charts.

To my knowledge. Flink does not do something like that for you, at least not yet :)

Bottom line is that you can use linux commands: iostat, dstat, htop, etc.

Hope this helps!

Andra

On Fri, Aug 7, 2015 at 5:07 PM, Stefanos Antaris <[hidden email]> wrote:

Hi to all.

I am working an a research project using flink and i would like to extract the CPU + RAM resources consumed on each worker in order to include the stats in my paper. Can anyone  advice me on how could i extract them?

Thanks in advance.
Stefanos

[Stefanos Antaris]

Hi to all.

Thank you for your quick response. I will try both of your solutions in order to find which one I the best.

Thanks,
Stefanos

On 7 Aug 2015 20:40, "Robert Metzger" <[hidden email]> wrote:

Hi Stefanos,

you can also write yourself a little script/tool which is periodically requesting the following JSON from the JobManager:

http://localhost:8081/setupInfo?get=taskmanagers&_=1438972693441

It returns a JSON string like this:

{"taskmanagers":[{"path":"akka:\/\/flink\/user\/taskmanager","dataPort":51695,"timeSinceLastHeartbeat":3,"slotsNumber":1,"freeSlots":1,"cpuCores":8,"physicalMemory":15941,"freeMemory":736,"managedMemory":461,"instanceID":"187cdbc1de0acdb02b5bdc796cefceaf","metrics":{"version":"3.0.0","gauges":{"cpuLoad":{"value":0.09425878320479862},"gc.PS-MarkSweep.count":{"value":6},"gc.PS-MarkSweep.time":{"value":764},"gc.PS-Scavenge.count":{"value":3},"gc.PS-Scavenge.time":{"value":121},"load":{"value":3.04},"memory.heap.committed":{"value":771751936},"memory.heap.init":{"value":805306368},"memory.heap.max":{"value":771751936},"memory.heap.usage":{"value":0.741875586302384},"memory.heap.used":{"value":572543920},"memory.non-heap.committed":{"value":50790400},"memory.non-heap.init":{"value":24576000},"memory.non-heap.max":{"value":318767104},"memory.non-heap.usage":{"value":0.09098903756392629},"memory.non-heap.used":{"value":29004312},"memory.pools.Code-Cache.usage":{"value":0.020928700764973957},"memory.pools.PS-Eden-Space.usage":{"value":0.1786344051361084},"memory.pools.PS-Old-Gen.usage":{"value":0.9994582533836365},"memory.pools.PS-Perm-Gen.usage":{"value":0.10412535071372986},"memory.pools.PS-Survivor-Space.usage":{"value":0},"memory.total.committed":{"value":822542336},"memory.total.init":{"value":829882368},"memory.total.max":{"value":1090519040},"memory.total.used":{"value":601962568}},"counters":{},"histograms":{},"meters":{},"timers":{}}}]}

On Fri, Aug 7, 2015 at 5:43 PM, Andra Lungu <[hidden email]> wrote:

Hi,

You can use something like iostat to extract the CPU usage.

For instance, I call this script on the JM node:
#!/usr/bin/env bash

lines=`cat /home/andra.lungu/hostnames.txt | paste -d, -s`
echo "${lines}"
IFS="," read -ra hostnames <<<"${lines}"
for line in "${hostnames[@]}"; do
  ssh ${line} `iostat -xtc 5 3 > /home/andra.lungu/cpu_stats/${line}` &
done

And then I parse the individual output in the cpu_stats folder with python to form charts.

To my knowledge. Flink does not do something like that for you, at least not yet :)

Bottom line is that you can use linux commands: iostat, dstat, htop, etc.

Hope this helps!

Andra

On Fri, Aug 7, 2015 at 5:07 PM, Stefanos Antaris <[hidden email]> wrote:

Hi to all.

I am working an a research project using flink and i would like to extract the CPU + RAM resources consumed on each worker in order to include the stats in my paper. Can anyone  advice me on how could i extract them?

Thanks in advance.
Stefanos