Skip to content

v3.5 Metrics Output

Andrey Kurilov edited this page Jul 12, 2017 · 12 revisions

Contents

  1. Load Average
    1.1. Console
    1.2. File
  2. Step Summary
    2.1. Console
    2.2. File
  3. I/O Traces
    3.1. Console
    3.2. File
  4. Threshold
    4.1. Console
    4.2. File

1. Load Average

Load average metrics records are thought to be produced periodically to monitor the load step state in the nearly real time mode. The producing period is configurable (output-metrics-average-period) and the default value is "10s" (10 seconds). Setting this period to 0 will disable the load average metrics output at all.

1.1. Console


  Step       |   Timestamp   |  Op  |Concurrency|       Count       | Step  |   Last Rate    |  Mean    |   Mean    
  Name       |               | Type |     x     |-------------------| Time  |----------------| Latency  | Duration  
(last 17 chars) yyyyMMdd-HHmmss Drivers Success Failed [s] [op/s] [MB/s] [us] [us]
16.239-1778629809 20170711-132716 CREATE 100x1 0 0 0.005 0.0 0.0 0 0
16.239-1778629809 20170711-132726 CREATE 100x1 105 0 10.006 8.801311 880.131 322077 7556263
16.239-1778629809 20170711-132736 CREATE 100x1 421 0 20.006 23.42035 2342.03 188145 4288254
16.239-1778629809 20170711-132746 CREATE 100x1 736 0 30.006 28.46382 2846.38 169945 3776626
16.239-1778629809 20170711-132756 CREATE 100x1 1057 0 40.006 30.66969 3066.96 161022 3569481
16.239-1778629809 20170711-132806 CREATE 100x1 1377 0 50.006 31.45123 3145.12 159180 3457120
16.239-1778629809 20170711-132816 CREATE 100x1 1699 0 60.006 32.03295 3203.29 156787 3393158
16.239-1778629809 20170711-132826 CREATE 100x1 2008 0 70.005 31.22738 3122.73 154213 3351128
16.239-1778629809 20170711-132836 CREATE 100x1 2326 0 80.006 31.36938 3136.93 152260 3349824
16.239-1778629809 20170711-132846 CREATE 100x1 2643 0 90.006 31.58480 3158.48 152738 3328149
Field Name Description
Step Name The configured step name. Automatic value is used if not configured obviously. Note that only last 17 characters are displayed in the table.
Timestamp The datetime of the record output in the "yyyyMMdd-HHmmss" format
Op Type Load operation type. Colored for readability.
Concurrency x Drivers The configured concurrency level and the count of the storage drivers used to perform a load. The effective concurrency level is a product of configured concurrency and the drivers count.
Count / Success The count of the items processed sucessfully.
Count / Failed The count of the items processed with a failure.
Step Time [s] The test step elapsed time in seconds. Note that the step elapsing more than 115 days will cause the cell overflow as far as only 7 characters are available for the output.
Last Rate / [op/s] The moving average operations per second rate for the last period (10 seconds by default).
Last Rate / [MB/s] The moving average megabytes per second rate for the last period (10 seconds by default).
Mean Latency [us] The last mean latency measured in the microseconds.
Mean Duration [us] The last mean operation duration measured in the microseconds.

1.2. Files

Average metrics data is written to a CSV file metrics.csv with dynamic path.

Field Name Description
DateTimeISO8601 Timestamp in the ISO8601 format
TypeLoad Load type (CREATE/READ/...)
Concurrency The concurrency level
DriverCount Count of the storage drivers used (may be more than 1 in the distributed mode)
CountSucc Total successful I/O tasks count
CountFail Total failed I/O tasks count
Size Total transferred byte count
JobDuration[s] Total step duration
DurationSum[s] Total sum of the I/O tasks durations
TPAvg[op/s] Total average throughput
TPLast[op/s] Last final moving average throughput
BWAvg[MB/s] Total average bandwidth
BWLast[MB/s] Last final moving average bandwidth
DurationAvg[us] Total average I/O tasks duration
DurationMin[us] Minimum I/O task duration
DurationLoQ[us] Low quartile of the I/O tasks duration distribution
DurationMed[us] Median of the I/O tasks duration distribution
DurationHiQ[us] High quartile of the I/O tasks duration distribution
DurationMax[us] Maximum I/O task duration
LatencyAvg[us] Total average I/O tasks latency
LatencyMin[us] Minimum I/O task latency
LatencyLoQ[us] Low quartile of the I/O tasks latency distribution
LatencyMed[us] Median of the I/O tasks latency distribution
LatencyHiQ[us] High quartile of the I/O tasks latency distribution
LatencyMax[us] Maximum I/O task latency

2. Step Summary

2.1. Console

2.2. Files

3. I/O Traces

The metrics for each load operation (request either file operation).

3.1. Console

The console output is absent.

3.2. Files

The file output is disabled by default. To enable the file output, set the output-metrics-trace-persist configuration parameter to "true". Output file: io.trace.csv with dynamic path.

Available fields

Field Name Description
StorageDriver The address/hostname of the storage driver executed the operation
StorageNode The target storage node address/hostname
ItemPath The resulting item path
IoTypeCode The I/O operation type code
StatusCode The I/O operation resulting status code
ReqTimeStart The I/O operation start timestamp in microseconds
Duration The I/O operation total duration in microseconds
RespLatency The I/O operation response latency in microseconds
DataLatency The I/O operation response data latency ("1st byte" of the response content) in microseconds
TransferSize The count of the bytes transferred within the I/O operation

IoTypeCode

Code I/O Type
0 NOOP
1 CREATE
2 READ
3 UPDATE
4 DELETE

StatusCode

Code Description HTTP response codes
O Pending (internal)
1 Active (internal)
2 Interrupted
3 Unknown failure all other codes
4 Success 2xx
5 I/O Failure
6 Timeout 504
7 Unrecognized storage response
8 Client failure or invalid request 100, 400, 405, 409, 411, 414, 416
9 Internal storage failure 413, 415, 429, 500, 501, 502, 503, 505
10 Item not found on the storage 404
11 Authentication/access failure 401, 403
12 Data item corrupted 2xx
13 Not enough space on the storage 507

4. Threshold

4.1. Console

4.2. Files

Clone this wiki locally