pm2/lib/Interactor/TransactionAggregator.js

416 lines
14 KiB
JavaScript

/**
* Copyright 2013 the PM2 project authors. All rights reserved.
* Use of this source code is governed by a license that
* can be found in the LICENSE file.
*/
/**
* Dependencies
*/
var cst = require('../../constants.js');
var log = require('debug')('pm2:aggregator');
var async = require('async');
var LABELS = {
"HTTP_RESPONSE_CODE_LABEL_KEY": 'http/status_code',
"HTTP_URL_LABEL_KEY": 'http/url',
"HTTP_METHOD_LABEL_KEY": 'http/method',
"HTTP_RESPONSE_SIZE_LABEL_KEY": 'http/response/size',
"STACK_TRACE_DETAILS_KEY": 'stacktrace',
"ERROR_DETAILS_NAME": 'error/name',
"ERROR_DETAILS_MESSAGE": 'error/message',
"HTTP_SOURCE_IP": 'http/source/ip',
"HTTP_PATH_LABEL_KEY": "http/path"
}
var units = { NANOSECONDS : 1 / (1000 * 1000), MICROSECONDS : 1 / 1000, MILLISECONDS : 1 };
units.SECONDS = 1000 * units.MILLISECONDS;
units.MINUTES = 60 * units.SECONDS;
units.HOURS = 60 * units.MINUTES;
units.DAYS = 24 * units.HOURS;
// EWMA = ExponentiallyWeightedMovingAverage from
// https://github.com/felixge/node-measured/blob/master/lib/util/ExponentiallyMovingWeightedAverage.js
// used to compute the nbr of time per minute that a variance is hit by a new trace
function EWMA(timePeriod, tickInterval) {
this._timePeriod = timePeriod || 1 * units.MINUTE;
this._tickInterval = tickInterval || 5 * units.SECONDS;
this._alpha = 1 - Math.exp(-this._tickInterval / this._timePeriod);
this._count = 0;
this._rate = 0;
var self = this;
this._interval = setInterval(function () {
self.tick();
}, this._tickInterval);
this._interval.unref();
};
EWMA.prototype.update = function(n) {
this._count += n || 1;
};
EWMA.prototype.tick = function() {
var instantRate = this._count / this._tickInterval;
this._count = 0;
this._rate += (this._alpha * (instantRate - this._rate));
};
EWMA.prototype.rate = function(timeUnit) {
return (this._rate || 0) * timeUnit;
};
/**
*
* # Data structure sent to interactor
*
* {
* 'process_name': {
* process : {}, // PM2 process meta data
* data : {
* routes : {
* '/' : [{ // deviance
* child : [
* ... // transactions
* ],
* count: 50, // count of this deviance
* max: 300, // max latency of this deviance
* min: 50, // min latency of this deviance
* mean: 120 // mean latency of this deviance
* }]
* },
* meta : {
* trace_nb : 50, // trace number
* mean_latency : 40, // global app latency in ms
* req_min : 30, // global app req per minutes
* db_trans_min : 20, // number of database transaction per min
* 500_count : 0, // number of 500 errors per min
* 404_count : 0, // number of 404 errors per min
* }
* }
* }
* }
*/
var TransactionAggregator = module.exports = function (pushInteractor) {
if (!(this instanceof TransactionAggregator))
return new TransactionAggregator(pushInteractor);
this.processes = {};
var FLUSH_INTERVAL = process.env.NODE_ENV === 'local_test' ? 1000 : 5000;
var self = this;
setInterval(function () {
var normalized = {};
// for every process
async.forEachOf(self.processes, function (process, name, next) {
var routes = process.routes;
normalized[name] = {
data : {
routes : {},
meta : {}
},
process : process.process
};
// for every route
async.forEachOf(routes, function (variances, route, next2) {
// get top 5 variances of the same route
variances = variances.sort(function (a, b) {
return a.count - b.count;
}).slice(0, 5);
normalized[name].data.routes[route] = [];
variances.forEach(function (variance) {
// deep copy
var newVariance = JSON.parse(JSON.stringify({
spans: variance.spans,
count: variance.count,
min: variance.min,
max: variance.max,
mean: variance.mean
}));
// replace meter object by his value
newVariance.meter = Math.round(variance.meter.rate(1 * units.SECONDS) * 100 ) / 100;
// delete stackrace from spans
self.applyOnTree(newVariance.spans, function (span) {
delete span.labels.stacktrace;
})
// push serialized into normalized data
normalized[name].data.routes[route].push(newVariance);
})
return next2();
}, next);
}, function () {
if (process.env.NODE_ENV === 'test') return ;
if (process.env.PM2_DEBUG) console.log(JSON.stringify(normalized));
// send the buffer to keymetrics
Object.keys(normalized).forEach(function(key) {
pushInteractor.bufferData('axm:transaction', normalized[key]);
});
})
}, FLUSH_INTERVAL);
this.aggregate = function (event, packet) {
if (!packet.data) return ;
if (!self.processes[packet.process.name])
self.processes[packet.process.name] = {
routes : {},
process : packet.process
};
var routes = self.processes[packet.process.name].routes;
if (!packet.data.traces)
return log('Got packet without traces : ' + JSON.stringify(Object.keys(packet.data)));
log('Aggregating %s new traces', packet.data.traces.length)
async.eachLimit(packet.data.traces, 1, function (trace, next) {
// convert spans list to trees
self.convertSpanListToTree(trace, function (tree) {
trace.spans = tree;
delete tree.labels.stackrace
// get the path from first span
var path = trace.spans.labels[LABELS.HTTP_PATH_LABEL_KEY];
// censor data in spans
self.censorTreeSpans(trace.spans);
self.matchPath(path, routes, function (matched) {
if (!matched) {
routes[path] = [];
log('Path %s isnt aggregated yet, creating new entry', path)
self.mergeTrace(routes[path], trace, next);
}
else {
log('Path %s already aggregated under %s, merging', path, matched)
self.mergeTrace(routes['/' + matched], trace, next);
}
})
})
}, function(error) {
if (error)
console.error(error);
});
}
this.mergeTrace = function (aggregated, trace, cb) {
self.computeSpanDuration(trace.spans)
var merge = function (variance) {
// no variance found so its a new one
if (!variance) {
delete trace.projectId;
delete trace.traceId;
trace.count = 1;
trace.mean = trace.min = trace.max = trace.spans.mean;
trace.meter = new EWMA();
trace.meter.update();
aggregated.push(trace);
}
// variance found, merge spans
else {
variance.min = variance.min > trace.spans.mean ? trace.spans.mean : variance.min;
variance.max = variance.max < trace.spans.mean ? trace.spans.mean : variance.max;
variance.mean = (trace.spans.mean + (variance.mean * variance.count)) / (variance.count + 1);
// update duration of spans to be mean
self.updateSpanDuration(variance.spans, trace.spans, variance.count, true);
variance.meter.update();
variance.count++;
}
return cb();
}
if (!aggregated)
return ;
// for every variance, check spans same variance
for (var i = 0; i < aggregated.length; i++) {
if (self.compareTree(aggregated[i].spans, trace.spans))
return merge(aggregated[i])
}
// else its a new variance
return merge(null);
}
this.convertSpanListToTree = function (trace, cb) {
var head, spans = trace.spans;
async.each(spans, function (current, next) {
if (current.parentSpanId == 0) {
head = current;
return next();
}
for (var i = 0, len = spans.length; i < len; i++) {
if (current.parentSpanId !== spans[i].spanId) continue ;
if (!spans[i].child) spans[i].child = [];
spans[i].child.push(current);
return next();
}
return next();
}, function () {
return cb(head);
});
}
/**
* Apply a function on all element of a tree
*/
this.applyOnTree = function(head, fn) {
fn(head);
if (head.child instanceof Array)
return head.child.forEach(fn);
}
/**
* Compute duration of a span from child key "startTime" and "endTime"
*/
this.computeSpanDuration = function (head) {
self.applyOnTree(head, function (span) {
if (span.endTime && span.startTime)
span.min = span.max = span.mean = Math.round(new Date(span.endTime) - new Date(span.startTime));
delete span.endTime;
delete span.startTime;
})
}
/**
* Parkour simultaneously both trees to update value of the first one using value of the second one
* The first should be variance already aggregated for which we want to merge the second one
* The second one is a new trace, so we need to re-compute mean/min/max time for each spans
*/
this.updateSpanDuration = function (ref_spans, spans, count) {
// head
if (ref_spans.parentSpanId === 0 || ref_spans.parentSpanId === "0") {
ref_spans.mean = Math.round((spans.mean + (ref_spans.mean * count)) / (count + 1) * 100 ) / 100;
ref_spans.min = ref_spans.min > spans.mean ? spans.mean : ref_spans.min;
ref_spans.max = ref_spans.max < spans.mean ? spans.mean : ref_spans.max;
}
// childs
if (!(ref_spans.child instanceof Array)) return ;
for (var i = 0, len = ref_spans.child.length; i < len; i++) {
var childspan = ref_spans.child[0];
childspan.mean = Math.round((spans.child[i].mean + (childspan.mean * count)) / (count + 1) * 100 ) / 100;
childspan.min = childspan.min > spans.child[i].mean ? spans.child[i].mean : childspan.min;
childspan.max = childspan.max < spans.child[i].mean ? spans.child[i].mean : childspan.max;
if (childspan.child instanceof Array)
self.updateSpanDuration(childspan, spans.child[i], count);
}
}
/**
* Compare a spans tree by going down on each span and comparing child and attribute
*/
this.compareTree = function (one, two) {
if (!one.child && !two.child) return true;
if (!one.child && two.child) return false;
if (one.child && !two.child) return false;
if (one.child.length !== two.child.length) return false;
for(var i = 0, len = one.child.length; i < len; i++) {
if (one.child[i].name !== two.child[i].name) return false;
if (one.child[i].kind !== two.child[i].kind) return false;
if (one.child[i].child)
return self.compareTree(one.child[i], two.child[i]);
}
return true;
}
/**
* Will return the route if we found an already matched route
*/
this.matchPath = function (path, routes, cb) {
var self = this;
// remove the last slash if exist
if (path[path.length - 1] === '/')
path = path.substr(0, path.length - 1)
// split to get array of segment
path = path.split('/').filter(function (item) {
return !item ? null : item;
});
// if the path has only one segment, we just need to compare the key
if (path.length === 1)
return routes[path[0]] ? cb(routes[path[0]]) : cb(null);
// check in routes already stored for match
async.forEachOfLimit(routes, 10, function (data, route, next) {
var segments = route.split('/').filter(function (item) {
return !item ? null : item;
});
if (segments.length !== path.length) return next(null);
for (var i = path.length - 1; i >= 0; i--) {
// different segment, try to find if new route or not
if (path[i] !== segments[i]) {
// case if the aggregator already have matched that path into a route and we got an identifier
if (self.isIdentifier(path[i]) && segments[i] === '*' && path[i - 1] === segments[i - 1])
return next(segments.join('/'));
// case a var in url match, so we continue because they must be other var in url
else if (path[i - 1] !== undefined && path[i - 1] === segments[i - 1] && self.isIdentifier(path[i]) && self.isIdentifier(segments[i])){
segments[i] = '*';
// update routes in cache
routes[segments.join('/')] = routes[route];
delete routes[route];
route = segments.join('/');
}
else
return next();
}
}
// if finish to iterate over segment of path, we must be on the same route
return next(segments.join('/'))
}, cb)
}
/**
* Check if the string can be a id of some sort
*/
this.isIdentifier = function (id) {
id = typeof(id) !== 'string' ? id + '' : id;
// uuid v1/v4 with/without dash
if (id.match(/[0-9a-f]{8}-[0-9a-f]{4}-[14][0-9a-f]{3}-[0-9a-f]{4}-[0-9a-f]{12}|[0-9a-f]{12}[14][0-9a-f]{19}/i))
return true;
// if number
else if (id.match(/\d+/))
return true;
// if suit of nbr/letters
else if (id.match(/[0-9]+[a-z]+|[a-z]+[0-9]+/))
return true;
else
return false;
}
var REGEX_VALUES_CLEAN = /":(?!\[|{)\\"[^"]*\\"|":(?!\[|{)[^,}\]]*|":\[[^{]*]/g;
/**
* Remove value from db queries before aggregation
*/
this.censorTreeSpans = function (head) {
self.applyOnTree(head, function (span) {
if (span.labels.results)
delete span.labels.results;
if (span.labels.result)
delete span.labels.result;
if (typeof(span.labels.cmd) === 'string')
span.labels.cmd = span.labels.cmd.replace(REGEX_VALUES_CLEAN, '\": \"?\"');
if (typeof(span.labels.sql) === 'string')
span.labels.sql = span.labels.sql.replace(REGEX_VALUES_CLEAN, '\": \"?\"');
if (typeof(span.labels.values) === 'string')
span.labels.values = span.labels.values.replace(REGEX_VALUES_CLEAN, '\": \"?\"');
if (typeof(span.labels.arguments) === 'string')
span.labels.arguments = span.labels.arguments.replace(REGEX_VALUES_CLEAN, '\": \"?\"');
})
}
};