Monitor Result

  1. Check whether performance statistics log files are generated.

    Check whether plugin performance statistics logs are generated in the directory specified by ps_log_dir.

    If ps_interval_time is set to 6, the program generates end-to-end log files, single-plugin log files, and throughput log files 6 seconds after the program is started.

    If ps_queue_size_interval_time is set to 50 and ps_queue_size_times is set to 100, the program generates a queue length log file 50 x 100 milliseconds (5 seconds) after startup. For details, see the following figure.

    Table 1 File types

    File Name

    File Type

    performance—statistics.log.e2e.20210329-183958.877032

    End-to-end log file

    performance—statistics.log.plugin.20210329-183958.877905

    Single plugin log file

    performance—statistics.log.queue.20210329-183957.893092

    Queue length log file

    performance—statistics.log.tpr.20210329-183958.883263

    Throughput log file

  2. Check the end-to-end log files.
    The following is an example of an end-to-end log file:
    {"average":"45138","frequency":"3","maxTime":"73058","minTime":"30511","streamName":"classification+detection","totalTime":"135416","type":"e2e","updateTime":"2021-03-29 20:13:59.717927"}
    Table 2 Log parameters

    Parameter

    Description

    average

    Average end-to-end duration (µs).

    frequency

    Number of end-to-end statistics collection times.

    maxTime

    Maximum end-to-end duration (µs).

    minTime

    Minimum end-to-end duration (µs).

    streamName

    Stream name in the pipeline.

    totalTime

    Total end-to-end duration (µs).

    type

    Statistics type. The end-to-end type is e2e.

    updateTime

    Time when the statistics are collected.

  3. Check the log files of a single plugin.
    The following is an example of a single plugin log file:
    {"average":"18546","elementName":"mxpi_modelinfer0","factory":"mxpi_modelinfer","frequency":"3","maxTime":"18608","minTime":"18493","streamName":"classification+detection","totalTime":"55638","type":"plugin","updateTime":"2021-03-29 20:13:59.720010"}
    Table 3 Log parameters

    Parameter

    Description

    average

    Average duration of a single plugin (µs).

    elementName

    Plugin element name.

    factory

    Plugin type.

    frequency

    Number of inference times processed by the plugin.

    maxTime

    Maximum duration (µs).

    minTime

    Minimum duration (µs).

    streamName

    Stream name in the pipeline.

    totalTime

    Total duration of a single plugin (µs).

    type

    Statistical type. For a single plugin, the statistical type is plugin. For model inference, the statistical type is modelInference. For post-processing, the statistical type is postProcessor.

    updateTime

    Time when the statistics are collected.

  4. Check the throughput log files.
    The following is an example of a throughput log file:
    {"streamName":"classification+detection","throughput":3,"throughputRate":0.5,"type":"throughputRate","updateTime":"2021-03-29 20:13:59.721819"}
    Table 4 Log parameters

    Parameter

    Description

    streamName

    Stream name in the pipeline.

    throughput

    Throughput

    throughputRate

    Throughput rate

    type

    Statistical type. The statistical type of throughput is throughputRate.

    updateTime

    Time when the statistics are collected.

  5. Check the queue length log files.

    The following is an example of a queue length log file:

    The queue length is always displayed as 0. Even during the inference process, the collected queue length is still 0. This indicates that the data is stored in the queue for a very short time during the inference process and is quickly obtained by the downstream plugin. As a result, the obtained queue length is 0. The queue length is not 0 only when congestion occurs in the inference process and data is stored in the queue.

    {"curSize":0,"elementName":"queue0","intervalTime":50000,"maxSize":0,"maxSizeBuffers":200,"recentSize":[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],"streamName":"classification+detection","type":"queueSize","updateTime":"2021-03-29 20:14:45.953312"}
    Table 5 Log parameters

    Parameter

    Description

    curSize

    Current queue length, that is, the last value in the recentSize list.

    elementName

    Plugin element name.

    intervalTime

    Statistics collection interval, which is the same as the value of ps_queue_size_interval_time.

    maxSize

    Maximum queue length during the statistics collection period, that is, the maximum value in the recentSize list.

    maxSizeBuffers

    Total queue length.

    recentSize

    List of queue lengths collected ps_queue_size_times at an interval of ps_queue_size_interval_time, in milliseconds.

    streamName

    Stream name in the pipeline.

    type

    Statistics type. The statistical type of the queue length is queueSize.

    updateTime

    Time when the statistics are collected.