hramos8 / mswyw

mswyw - Microservice: Worth Your Weight?

mswyw - Microservice: Worth Your Weight?

What it is

A little utility that gathers some metrics about your deployed services and computes 1 metric - mswyw. It quantifies if your microservice is worth its weight. The formula is simple:

mswyw = VALUE / COST

Examples of VALUE attributes:

• Number of endpoints (presumably the more endpoints, the more functionality, the more useful. Controversial)
• Number of Use Case Points it implements (a measure of features/value to the user)
• Number of Function Points it implements (a measure of features/value to the user)
• APDEX - a measure of user satisfaction with the performance of your (micro)service
• Error Rate (epm - errors per minute) - a measure of user lack of satisfaction with your service
• Requests per minute (rpm) - the more your service can take, the leaner it is to run.

Examples of COST attributes:

• Total amount of RAM used by all replicas of your (micro)service. You know that Amazon charges for that, right? The less memory you need, the cheaper the machine instance you can get away with. It impacts you recurring expense.
• Amount of CPU percentage used by each of your replicas. You also pay for CPU. A smaller machine with optimized/faster code will cost you less. Also a recurring expense on the cloud.

Formulas

You can plug your own formula by passing --calcProvider=aPythonModule . Your module should implement 1 simple python function (calc_mswyw). We provide a default implementation in module formula.py (default python module used): mswyw = a * [( b * apdex + c * rpm + d * endpoints) / (e * mem + f * cpu + g * epm)]

Note that a,b,c etc are just coefficients which you can override passing --coefficients as a json, with these key names for the coefficients:

• a: "total"
• b: "apdex"
• c: "rpm"
• d: "endpoints"
• e: "mem"
• f: "cpu"
• g: "epm"

Don't worry, we provide default coefficient values as well. But you can tweak when you want. For example, use 0.0 for a coefficient to kick that element out of the formula (say "I don't want number of endpoints to have any influence on it" - pass "endpoints":0.0).

Again, if you want a totally different formula just take a look at formula.py and implement your own module, add it to PYTHONPATH and pass it in with --calcProvider.

Motivation

Use and abuse of microservices is intense lately. I have seen my fair share of convoluted, bloated Dockerfiles, not to mention REST microservices that eat 300MB RAM and with one single endpoint, which tells if a person's document ID is well formed. You know, the kind of thing you do with 20 lines of code. This kind of utility is much better suited for a reusable library. If it is to be shared by many projects... sure, go ahead and make it into a separate git repo and produce a binary JAR (if in Java) which people can import. It does not need to be a microservice.

I agree with Bob Martin that one should be able to defer the deploy decision. I should be able to deploy code with a REST façade of 20 endpoints or as 2 services each with a REST façade of 10 endpoints each. Or with a SOAP façade. Or both. Or GraphQL. You get the idea - don't hardcode the façade technology into your reusable core. Read about clean architectures, for example.

By obtaning a metric we hope to let people compare the "fat factor" (worth your weight) of their (micro)services.

How to install

Make sure your Python has paver installed (pip3 install paver) and then run:

pip3 install git+https://github.com/sglebs/mswyw

How to run

You need to tell mswyw how to gather runtime information (RAM, APDEX etc) by informing the name of a provider, in the form of a fully qualified name of a Python module that implements the API we need. If, instead, the value you provide is a literal json, we will use that instead. If the value is a valid file path, we assume it is a json file with the values we need.

Currently the runtime providers supported are New Relic (--runtimeProvider=nrelic) and Elastic APM (--runtimeProvider=elastic).

Example/NewRelic:

Remember: you need the paid version. The free version does not support the APIs we call.

mswyw --runtimeProvider=nrelic --providerParams={"nrelic.APPID":"123456","nrelic.APIKEY":"ABCDEFG"}

You can also get consolidated results for a collection of apps by name (regex) if you prefer:

mswyw --runtimeProvider=nrelic --providerParams={"nrelic.APPS":"foo.*bar$","nrelic.APIKEY":"ABCDEFG"}

ElasticAPM via Elastic indices / Example:

APDEX calculation depends on the "T" value in seconds via "elastic.APDEX_T" (the default is 0.5 seconds). It is needed so we can compute the APDEX on Elastic data.

mswyw --runtimeProvider=elastic --providerParams={"elastic.APPS":"foo", "elastic.URL":"http://elastic.softplan.com.br:9200", "elastic.USER":"myUser", "elastic.PASSWORD":"myPassword", "elastic.APDEX_T": 2 }

The example above overrides the default value for APDEX_T.

How to fail a build pipeline

If you want to fail a build pipeline based on the mswyw score you can use the --minResult parameter. For example, if the mswyw score should be >= 0.5, you should pass --minResult=0.5 You may want to do this while running stress tests, so you can evaluate how your microservice behaves.

Special Thanks

We would like to thank Softplan for supporting the development of this utility.