MRANGE returning out-of-query results

[dpipemazo]

Hi all -- originally posted this issue on the grafana datasource but I think it might fit better here: RedisGrafana/grafana-redis-datasource#56

I'm using Redis Time Series to monitor per-process CPU status on my machine, using psutil to scrape the system for process names and then monitor their CPU percentage over the sampling interval to unique, per-process keys with shared filters. I'm then using MRANGE to query and plot the data in grafana.

This sort of time-series data is inherently sparse -- over a given time window there will be processes that exist in the overall query set across the filters (i.e. all processes ever) but that don't have any data within the time window. I'm seeing some interesting results across an example query:

ts.mrange 1598405335639 1598410333697 AGGREGATION avg 1000 WITHLABELS FILTER element=monitor type=process subtype1=cpu_user

Full query response here:

mrange.txt

A few examples of results that are out-of-query, either happening before the start timestamp or after the stop timestamp:

1484) 1) "monitor:process:99:kworker/7:1:cpu_user"
      2) 1) 1) "element"
            2) "monitor"
         2) 1) "type"
            2) "process"
         3) 1) "container"
            2) "x86_64"
         4) 1) "device"
            2) "default"
         5) 1) "language"
            2) "Python"
         6) 1) "version"
            2) "1.7.2"
         7) 1) "level"
            2) "INFO"
         8) 1) "subtype0"
            2) "99:kworker/7:1"
         9) 1) "subtype1"
            2) "cpu_user"
      3) 1) 1) (integer) 1598456570000
            2) 0

1482) 1) "monitor:process:9866:kworker/3:0:cpu_user"
      2) 1) 1) "element"
            2) "monitor"
         2) 1) "type"
            2) "process"
         3) 1) "container"
            2) "x86_64"
         4) 1) "device"
            2) "default"
         5) 1) "language"
            2) "Python"
         6) 1) "version"
            2) "1.7.2"
         7) 1) "level"
            2) "INFO"
         8) 1) "subtype0"
            2) "9866:kworker/3:0"
         9) 1) "subtype1"
            2) "cpu_user"
      3) 1) 1) (integer) 1598404695000
            2) 0

This is running Redis Time Series 1.2.7. I did see a bug (#464) that addresses something similar for RANGE but am not sure if it would apply to MRANGE. I'll try upgrading to the latest version and will report on if I can continue to reproduce.

[dpipemazo]

Upgraded Redis Time Series to latest version (commit e8030d14e7b8e186518b1b551cfa3cde261bf3c3) and the behavior has improved here -- no longer seeing the issue. Thanks!