diff options
Diffstat (limited to 'site/src/main/jekyll/core/metrics.md')
-rw-r--r-- | site/src/main/jekyll/core/metrics.md | 80 |
1 files changed, 65 insertions, 15 deletions
diff --git a/site/src/main/jekyll/core/metrics.md b/site/src/main/jekyll/core/metrics.md index ab144d13..dcef8304 100644 --- a/site/src/main/jekyll/core/metrics.md +++ b/site/src/main/jekyll/core/metrics.md @@ -12,32 +12,82 @@ Philosophy ---------- Back in the day, the most common approach to get metrics out of an Akka/Spray application for production monitoring was -doing manual instrumentation: select your favorite metrics collection library, wrap you messages with some useful metadata, -wrap your actor's receive function with some metrics measuring code and, finally, push that metrics data out to somewhere -you can keep it, graph it and analyze it whenever you want. +doing manual instrumentation: select your favorite metrics collection library, wrap you messages with some useful +metadata, wrap your actor's receive function with some metrics measuring code and, finally, push that metrics data out +to somewhere you can keep it, graph it and analyse it whenever you want. -Each metrics collection library has it's own strengths and weaknesses, and each developer has to choose wisely according to the -requirements they have in hand, leading them in different paths as they progress with their applications. Each path has -different implications with regards to introduced overhead and latency, metrics data accuracy and memory consumption. Kamon takes this -responsibility out of the developer and tries to make the best choice to provide high performance metrics collection instruments -while keeping the inherent overhead as low as possible. +Each metrics collection library has it's own strengths and weaknesses, and each developer has to choose wisely according +to the requirements they have in hand, leading them in different paths as they progress with their applications. Each +path has different implications with regards to introduced overhead and latency, metrics data accuracy and memory +consumption. Kamon takes this responsibility out of the developer and tries to make the best choice to provide high +performance metrics collection instruments while keeping the inherent overhead as low as possible. Kamon tries to select the best possible approach, so you don't have to. - - Metrics Collection and Flushing ------------------------------- -All the metrics infrastructure in Kamon lives around two concepts: collection and flushing. Metrics collection happens in real time, as soon -as the information is available for being recorded. Let's see a simple example: as soon as a actor finishes processing a -message, Kamon knows the elapsed time for processing that specific message and it is recorded right away. If you have millions -of messages passing through your system, then millions of measurements will be taken. +All the metrics infrastructure in Kamon was designed around two concepts: collection and flushing. Metrics collection +happens in real time, as soon as the information is available for being recorded. Let's see a simple example: as soon as +a actor finishes processing a message, Kamon knows the elapsed time for processing that specific message and it is +recorded right away. If you have millions of messages passing through your system, then millions of measurements will be +taken. Flushing happens recurrently after a fixed amount of time has passed, a tick. Upon each tick, Kamon will collect all measurements recorded since the last tick, flush the collected data and reset all the instruments to zero. Let's explore a little bit more on how this two concepts are modeled inside Kamon. -SIMPLE CLASS DIAGRAM WITH COLLECTION AND FLUSHING SIDES +<img class="img-responsive" src="/assets/img/diagrams/metric-collection-concepts.png"> + +A metric group contains various individual metrics that are related to the same entity, for example, if the entity we +are talking about is an actor, the metrics related to processing time, mailbox size and time in mailbox for that +specific actor are grouped inside a single metric group, and each actor gets its own metric group. As you might disguise +from the diagram above, on the left we have the mutable side of the process that is constantly recoding measurements as +the events flow through your application and on the right we have the immutable side, containing snapshots representing +all the measurements taken during a specific period on time for a metric group. + + +Filtering Entities +------------------ + +By default Kamon will not include any entity for metrics collection and you will need to explicitly include all the +entities you are interested in, be it a actor, a trace, a dispatcher or any other entity monitored by Kamon. The +`kamon.metrics.filters` key on your application's configuration controls which entities must be included/excluded from +the metrics collection infrastructure. Includes and excludes are provided as lists of strings containing the +corresponding GLOB patterns for each group, and the logic behind is simple: include everything that matches at least one +`includes` pattern and does not match any of the `excludes` patterns. The following configuration file sample includes +the `user/job-manager` actor and all the worker actors, but leaves out all system actors and the `user/worker-helper` +actor. + +``` +kamon { + metrics { + filters = [ + { + actor { + includes = [ "user/job-manager", "user/worker-*" ] + excludes = [ "system/*", "user/worker-helper" ] + } + }, + { + trace { + includes = [ "*" ] + excludes = [] + } + } + ] + } +} +``` + +Instruments +----------- + +Talk about how HDR Histogram works and how we use it. + + +Subscription protocol +--------------------- +Explain how to subscribe for metrics data and provide a simple example. |