Promql

Cheat sheet and labs

Data Models and Lablse

• Metrics Names Specyige the general feature of system taht is mesaured(http_request_toatl)
• Metrics Lables Enable Prometheus dimensional data modle to ideentyfie any given combination of labes for the same metiric anem
  • Can’t use __ resrved for the internal testing

DataTypes

• Stirng
• Scalar int
• Vectors
• Instant vector (One value )
  • Every 30s [30]
• Range vectors(Multiples values)
  • From yesterday to now

• Counter (sotres a metric that will increase over time and total amount of them)

  • If metrics is a total then this is counter
    • Example: alter_menager_recived_total
  • shoudln’t be used for decrease
  • Returns Instant Vector
  • Examples request count or error count

  total number of the metrics over desired period bash rate(counter_name [last_period_of_time])

  • increase (how much a counter has gone up over a certain period of time)
    • Imagine you have a counter for disk flush requests: At the start of the day, the counter is at 100. By the end of the day, it’s at 300. The increase is 200
  • Gauge (metrics that go up and down)
  • Cpu Usage,memory usage ,response time
  • Usually an input to the agregaitng functions like avrage over time

avg_over_time(http_response_time[5m])

• Hisotgram put data in a specyfic buckets(has to be predefined)
  • Stores total number of metrics
    • Sum of values
  • If metrics is a bucket then this is histogram
    • Example: prometheus_http_response_size_bytes_bucket

  • Better visualisation

    • predefined functions for procentails

    Hisotgram_qunatilne(0.95,sum(rate(request_duration_buckets))by(le))
    

    - Summery(if u don’t know the exact value of the buckets) - Response times - Respnse size - Bytes exchanged - ReadIO ## Filters - Based on labels

Rates and Derivities

rate() Stable increase of vector irate() The same as vecotrs unless spikes derive desinged to show changes more for an aletr then the daschboard

Cpu usage over the last cpu

sum by(cpu)(rate(node_cpu_seconds_total{mode!="idle"}[5m]))*100

sum by(path)(rate(http_request_total{status="500"} [1d]))

Aggregation over dimmesions

• lables add dimension to the dataset http_request_total has 3 lables application instance grouop**
• Get all the labes

sum(http_request_total)

• Exclude instaces

sum(http_request_total) without(instance)

• or based on another group

sum by (application_group)(http_request_total)

Timestamps Metrics

timestamps are given in epoch time and can be queried for instant vector - How fresh the data is

# So this will give u the raw data to 
timestamp(go_memstat_frees_total)

# To get human readable time use 
time() - timestamp(go_memstat_frees_total)

Base units