OpenMetrics: Prometheus Unbound?

The State of the Art of OpenMetrics and some fundamental novelties about it.

What it is? Why do we need it? In case we really need it.

Cloud_Native Rejekts EU 2019 - Barcelona, Spain - May 19th 2019.

Transcript

1. OpenMetrics.
   Prometheus Unbound?
   Leonardo Di Donato.
   Open Source Engineer @ Sysdig.
   Maintainer of Falco.
   Co-creator of kubectl-trace.
   Creator of go-syslog.
   @leodido
   2019.05.19 - Cloud_Native Rejekts EU - Barcelona, Spain
   

   View Slide

2. What OpenMetrics is? @leodido
   gh:openmetrics/openmetrics
   Just a wire protocol.
   A lingua franca in the whole observability story
   being built using the Prometheus exposition format as
   the starting point.
   It’s (still) a draft
   In the CNCF sandbox
   

   View Slide

3. Prometheus won the battle!
   Against no one ➡ there was hardly anyone in
   the landscape …
   Everyone knows it nowadays.
   It displays metrics line-by-line in a
   text-based format.
   Histograms, counters, gauges, summaries.
   N-dimensional space with labels.
   It is really good at doing metrics
   And what else?! ‍♂
   So why do we really need OpenMetrics?
   Mess
   
   Prometheus
   
   OpenMetrics
   @leodido
   

   View Slide

4. Standardization
   Traditional vendors want to avoid lock-in.
   Goal ➡ IETF RFC
   Secret goal ➡ kill the concept of hierarchical data models
   Open Standard:
   • no brand.
   • vendor neutral.
   • available to all to read and implement ➡ foster cooperation.
   OpenMetrics
   @leodido
   

   View Slide

5. Novelties
   Mostly the same of Prometheus except ...
   UNIX timestamps in seconds
   What are you counting?
   # UNIT foo_seconds seconds ⬅
   not applicable to state sets and info metrics
   UNIT descriptor directive
   # HELP test_m Bla bla bla description no one really reads
   # TYPE test_m counter
   # UNIT potatoes
   test_m{...} x
   test_m{...} x
   # EOF ⬅
   Metric sets followed by an end marker
   attach an ID off a trace to directly link to that trace
   foo_bucket{le="0.1"} 8 # {} 0.054
   foo_bucket{le="1"} 10 # {id="9856e"} 0.67
   foo_bucket{le="10"} 17 # {id="12fa8"} 9.8 1520879607.789
   ⬆
   Exemplars
   Registered content-type/mime-type
   Normalization
   • same escape rules for HELP directive and labels
   • better handling of white spaces between tokens
   Maybe application/openmetrics-text?
   @leodido
   

   View Slide

6. New metric
   types
   Mostly the same of Prometheus except ...
   # TYPE x info
   x_info{entity="ctrl",name="pretty",version="8.1"} 1
   x_info{entity="repl",name="prettier",version="8.2"} 1
   Info metrics
   Just histograms for gauges.
   What else am I supposed to say?
   Gauge histograms
   # TYPE foo stateset
   foo{entity="controller",foo="a"} 0
   foo{entity="controller",foo="b"} 1
   foo{entity="replica",foo="a"} 0
   foo{entity="replica",foo="b"} 1
   # EOF
   State sets
   What Prometheus calls untyped metrics.
   Unknown
   @leodido
   

   View Slide

7. State of the Art.
   Prometheus Python Client >= 2.5.0
   Do you want to verify your output is valid OpenMetrics?
   Google, Uber, OpenCensus ...
   ~2 years in the making, nothing more than this out there yet …
   Suggestions
   ● Better outward communication
   ● Test suite to let other developers build against it
   Give birth to it!
   @leodido
   

   View Slide

8. Thanks.
   Reach me out @leodido on twitter & github!
   SEE Y’ALL AROUND IN KUBECON
   http://bit.ly/openmetrics_bcn
   

   View Slide