...
There are two basic modes that can be run: db off and db on. db off has cache running in memory. db on uses the db for caching. When db is set to on, the nodesum files are cached into the db and filters are run against the db instead of doing the filters manually in memory. NMISx db caching is now the faster option. By default this config item is turned off, to enable this edit your opCommon.nmis file located at /usr/local/omk/conf/opCommon.nmis and set nmisx_db_cache_enable to 1. For systems with a low number of nodes, using the db may not show any performance improvement
| Code Block |
|---|
'nmisx_db_cache_enable' => 1 |
There are two more configuration items that control how the cache works:
| Code Block |
|---|
'nmisx_cache_min_age_before_refresh' => 60 |
This setting tells the app how long to wait before checking to see if the cache needs refreshing. In order to know if the cache needs refreshing we check the time the nodesum files are written and compare that against the time the cache was built. This is a simple optimisation. Since we access the cache quite often it gets a little expensive to check it several times a second, this prevents those checks for at least X seconds. The idea is that NMIS only writes the nodesum files at most once a minute so don’t bother looking more than that often.
| Code Block |
|---|
'nmisx_cache_time' => 900 |
This setting controls how long the data will stay in the cache before it’s removed. This is only for db mode, in memory mode this will have no affect.
...
To enable sorting of 8h health please edit opCommon.nmis and change:
| Code Block |
|---|
'nmisx_db_cache_node_health_sorting' => 1 |
...
Primary Poller Caveat
When using DB caching on a master Primary server, 8h and 16h summary data metrics for nodes not polled by the Primary are unable to be cached. In the node list you will find the 8h node summary metric missing.
pollers is not cached, example is showing the 8h health value in the nodes list. If this data is important we recommend disabling db caching for this your environment.
Subnet Caching
In environments with high node counts subnet generation can be expensive to process. To make sure the application remains fast we introduced a cache for this data and by default it has a 5min expiry.
The subnet generation Generation of this cache takes part in the web request while viewing a subnet map and the cache mtime is greater than the nmisx_subnet_cache_time. In instances with large node counts this can slow down a the users web request each time the cache needs to be regenerated, if needed you can adjust increase the cache time to and lessen the impact on users.
| Code Block |
|---|
'nmisx_subnet_cache_time' => 300 |
Background Subnet Caching
Recommended for larger node counts.
To perform subnet caching in the background you can use a cron job to call the opCharts CLI and perform a subnet cache caching manually, the interevile to run this is a period in which this is run is balance between not caching stale data and processing time it takes taken to generate subnets with performance considerations included, we would reccomend calculate subnets. We would recommend to run the cron job every hour. If you have implemented this method you should change nmisx_subnet_cache_time to 86400 so it will not be performed in a web request. which will stop the web thread from calculating subnets . An example cron job provided
| Code Block | ||||
|---|---|---|---|---|
| ||||
59 * * * * /usr/local/omk/bin/ | ||||
| Code Block | ||||
opcharts-cli.plexe act=refresh-subnet-cache >/dev/null 2>&1 |
Multiple web users
If you have multiple users of the platform you can increase the worker count, default is 10
| Code Block |
|---|
'omkd_workers' => 10 |