statshub is a repository for incrementally calculated statistics stored using a dimensional model.
Stats are updated and queried using a RESTful API.
statshub for Lantern is currently deployed to Heroku at http://pure-journey-3547.herokuapp.com/.
Every stat tracked by statshub is associated to a string id. When stats are updated, they are updated relative to their original values as tied to that string id.
statshub can roll up stats to any number of unrelated dimensions.
statshub tracks two kinds of stats:
Counters - these keep incrementing forever (e.g. an odometer)
Gauges - these track an absolute value that can change over time (a speedometer). Statshub assumes that gauges are reported every 5 minutes. It stores them in 6 minute buckets. For rolled up gauges, in order to avoid presenting incomplete or fluctuating information, the reported values reflect the prior 6 minute bucket. Consequently, they can be up to 6 minutes out of date.
Stats can be updated in one of four ways:
Counters - directly sets the value of a counter.
Increments - increments the existing value of a counter by some delta.
Gauges - directly sets the value of a gauge.
Members - tracks the value's membership in a set of unique values. The corresponding gauge value is calculated as the count of unique members.
MultiMembers - like Members, but allows submitting multiple members at once instead of just one.
Stats are queried at the dimension level. A query can ask for only a single dimension, or omit the dimension and receive stats for all dimensions.
For use by Lantern specifically, the statshub REST api caches queries for the "country" dimension for 1 minute. It does this as a performance optimization for this very frequent query.
Query results include a couple of special items:
Totals - for every dimension, statshub returns the total across all dimension keys.
GaugesCurrent - In addition to returning the usual Gauges (which are up to 6 minutes out of date), the query result also includes GaugesCurrent, which shows gauges for the current period. These are not particularly reliable because of the reasons mentioned above, but they can be handy for testing to make sure that updates are being recorded.
statshub archives its stats to Google Big Query every 10 minutes. It
authenticates using OAuth and connects to a specific project, using the
environment variables GOOGLE_PROJECT and GOOGLE_TOKEN.
statshub expects Google Big Query to contain a dataset named "statshub". It populates one table per dimension inside this dataset.
This example session submits and queries stats for the ids "myid1" and "myid2".
Here we are submitting and querying stats for the id myid1.
curl --data-binary \
'{"dims": {
"country": "es",
"user": "bob"
},
"counters": { "counterA": 50 },
"increments": { "counterB": 500 },
"gauges": { "gaugeA": 5000 },
"members": { "gaugeB": "item1" },
"multiMembers": { "gaugeC", ["itemI", "itemII"] }
}' \
"http://localhost:9000/stats/myid1"{"Succeeded":true,"Error":""} curl "http://localhost:9000/stats/" | python -mjson.tool{
"Error": "",
"Succeeded": true,
"dims": {
"country": {
"es": {
"counters": {
"counterA": 50,
"counterB": 500
},
"gauges": {
"gaugeB": 1
},
"gaugesCurrent": {
"gaugeA": 5000
}
},
"total": {
"counters": {
"counterA": 50,
"counterB": 500
},
"gauges": {
"gaugeB": 1,
"gaugeC": 2
},
"gaugesCurrent": {
"gaugeA": 5000
}
}
},
"user": {
"bob": {
"counters": {
"counterA": 50,
"counterB": 500
},
"gauges": {
"gaugeB": 1
},
"gaugesCurrent": {
"gaugeA": 5000
}
},
"total": {
"counters": {
"counterA": 50,
"counterB": 500
},
"gauges": {
"gaugeB": 1,
"gaugeC": 2
},
"gaugesCurrent": {
"gaugeA": 5000
}
}
}
}
}curl --data-binary \
'{"dims": {
"country": "es",
"user": "bob"
},
"counters": { "counterA": 60 },
"increments": { "counterB": 600 },
"gauges": { "gaugeA": 6000 },
"members": { "gaugeB": "item2" },
"multiMembers": { "gaugeC", ["itemI", "itemIII"] }
}' \
"http://localhost:9000/stats/myid2"{"Succeeded":true,"Error":""} curl "http://localhost:9000/stats/country" | python -mjson.tool{
"Error": "",
"Succeeded": true,
"dims": {
"country": {
"es": {
"counters": {
"counterA": 60,
"counterB": 1100
},
"gauges": {
"gaugeB": 2,
"gaugeC": 3
},
"gaugesCurrent": {
"gaugeA": 6000
}
},
"total": {
"counters": {
"counterA": 60,
"counterB": 1100
},
"gauges": {
"gaugeB": 2,
"gaugeC": 3
},
"gaugesCurrent": {
"gaugeA": 6000
}
}
}
}
}REDIS_ADDR=<host:port> REDIS_PASS=<password> GOOGLE_PROJECT=<project id> GOOGLE_TOKEN=<json encoded oauth config from oauther> PORT=9000 go run statshub.goNeed to configure the Redis address and password only once (these are persistent settings in Heroku).
heroku config:set REDIS_ADDR=<host:port>
heroku config:set REDIS_PASS=mR0bKNfhlxoKIHqnBA53
heroku config:set ARCHIVE_TO_BIGQUERY=true
heroku config:set GOOGLE_PROJECT=<project id>
heroku config:set OAUTH_CONFIG=<json encoded oauth config from oauther>You'll need to connect your git repo to Heroku first:
cd $GOPATH/src/github.com/getlantern/statshub
git remote add heroku git@heroku.com:pure-journey-3547.gitTo deploy:
git commit -a -m"..." && git push
git push heroku masterSometimes, it is necessary to delete keys from the database (e.g. to clean out corrupted/test data). This can be done using a LUA script.
For example, let's say that we need to delete all keys containing "traversalSucceeded". Running this command inside redis-cli will do that:
EVAL "return redis.call('del', unpack(redis.call('keys', ARGV[1])))" 0 "*traversalSucceeded*"
See StackOverflow for more details.