chore: Modernize the MongoDB Atlas Mixin #1544
Conversation
| - alert: MongoDBAtlasElectionTimeouts | ||
| annotations: | ||
| description: The number of elections being called due to the primary node timing out in replica set {{$labels.rs_m}} in cluster {{$labels.cl_name}} is {{printf "%.0f" $value}} which is above the threshold of 10. | ||
| description: The number of elections being called due to the primary node timing out in replica set {{$labels.rs_nm}} in cluster {{$labels.cl_name}} is {{printf "%.0f" $value}} which is above the threshold of 10. |
There was a problem hiding this comment.
Identified a typo here
Dasomeone
left a comment
Dasomeone
left a comment
There was a problem hiding this comment.
Have to leave it here as it's end of day, but first pass review sort of done.
Generally, layout is A+ and I'm perfectly happy with it.
Couple suggestions for improvements in terms of legend tabels and filtering, and I have yet to do a pass on the usage of common-lib so no comments there yet
| dashboardTimezone: 'default', | ||
| dashboardRefresh: '1m', | ||
| // Basic filtering - MongoDB Atlas uses job and cl_name (cluster name) as primary filters | ||
| filteringSelector: 'job="integrations/mongodb-atlas"', |
There was a problem hiding this comment.
As we've recently talked about, we can vendor latest logs-lib and unset this for the public mixin
| alertsDeadlocks: 10, // count | ||
| alertsSlowNetworkRequests: 10, // count | ||
| alertsHighDiskUsage: 90, // % | ||
| alertsSlowHardwareIO: 3, // seconds |
There was a problem hiding this comment.
Like I commented on a previous PR, we could consider having the units be more tightly coupled with the native metric unit, e.g. milliseconds in order to simplify the query
| { | ||
| alert: 'MongoDBAtlasSlowHardwareIO', | ||
| expr: ||| | ||
| (sum without (disk_name) (increase(hardware_disk_metrics_read_time_milliseconds[5m])) + sum without (disk_name) (increase(hardware_disk_metrics_write_time_milliseconds[5m]))) / 1000 > %(alertsSlowHardwareIO)s |
There was a problem hiding this comment.
Like I commented on a previous PR, we could consider having the units be more tightly coupled with the native metric unit, e.g. milliseconds in order to simplify the query
| hardwareIO: | ||
| commonlib.panels.generic.timeSeries.base.new('Hardware I/O', targets=[ | ||
| signals.cluster.diskReadCount.asTarget(), | ||
| signals.cluster.diskWriteCount.asTarget(), | ||
| ]) | ||
| + g.panel.timeSeries.panelOptions.withDescription("The number of read and write I/O's processed.") | ||
| + g.panel.timeSeries.standardOptions.withUnit('iops') | ||
| + g.panel.timeSeries.options.legend.withPlacement('right') | ||
| + g.panel.timeSeries.options.legend.withAsTable(true), | ||
|
|
||
| hardwareIOWaitTime: | ||
| commonlib.panels.generic.timeSeries.base.new('Hardware I/O wait time / $__interval', targets=[ | ||
| signals.cluster.diskReadTime.asTarget() | ||
| + g.query.prometheus.withInterval('2m'), | ||
| signals.cluster.diskWriteTime.asTarget() | ||
| + g.query.prometheus.withInterval('2m'), | ||
| ]) | ||
| + g.panel.timeSeries.panelOptions.withDescription('The amount of time spent waiting for I/O requests.') | ||
| + g.panel.timeSeries.standardOptions.withUnit('ms') | ||
| + g.panel.timeSeries.options.tooltip.withSort('desc') | ||
| + g.panel.timeSeries.options.legend.withPlacement('right') | ||
| + g.panel.timeSeries.options.legend.withAsTable(true), |
There was a problem hiding this comment.
For these two panels, I think it'd be beneficial if we make use of the table legend options to add last*, min, mean, and max columns here. Should be available via standard options, can't remember off the top of my head, but Gabriel just used it in the postgres mixin last week
| + g.panel.timeSeries.standardOptions.withUnit('reqps') | ||
| + g.panel.timeSeries.options.tooltip.withSort('desc'), | ||
|
|
||
| networkThroughput: |
There was a problem hiding this comment.
+1 for last*, and at least mean as additional data columns for a quick overview on the side
| // | ||
| // Elections panels | ||
| // |
| + g.panel.timeSeries.standardOptions.withUnit('reqps') | ||
| + g.panel.timeSeries.options.tooltip.withSort('desc'), | ||
|
|
||
| slowNetworkRequestsPerformance: |
There was a problem hiding this comment.
+1 panel here and general networkThroughputPerformance as well that may need some filtering as it is and will keep affecting the y axis scaling
| + g.panel.timeSeries.options.legend.withPlacement('right') | ||
| + g.panel.timeSeries.options.legend.withAsTable(true), | ||
|
|
||
| hardwareIOWaitTime: |
There was a problem hiding this comment.
+1 filtering, I'll stop commenting on them now
2 participants
Some of the screenshots are missing data mostly due to me not setting up sharding, but queries/functionality should be pretty similar to the original.
MongoDB Atlas cluster overview
Paginated the tables 🚀
MongoDB Atlas elections overview
MongoDB Atlas operations overview
MongoDB Atlas performance overview
MongoDB Atlas sharding overview
