-
Notifications
You must be signed in to change notification settings - Fork 57
vayacondios
Simple enough to use in a shell script, performant enough to use everywhere. Why the hell wouldn't you record that metric, ese?
You can chuck the following into vayacondios from ruby, the shell, over HTTP, or in a repeated polling loop:
- value -- 'there are 3 cups of coffee remaining':
- timing -- 'this response took 100ms'
- count -- 'I just served a 404 response code'
-
fact -- 'here's everything to know about the coffee machine:
{"cups_remaining":3,"uptime":127,room:"hyrule"}
' -- an arbitrary JSON hash
- Decentralized: Any (authorized) system can dispatch facts or metrics using an arbitrary namespace and schema. Nothing needs to be created in advance.
- Bulletproof: UDP clients will never fail because of network loss or timeout.
- Fast: UDP clients are non-blocking and happily dispatch thousands of requests per second.
- Minimal Dependency: Ruby clients uses nothing outside of the standard libraries.
- Ubiquitous: Can send facts from the shell (with nothing besides , from ruby, from
- writes are simple, reads are clever: A writer gets to chuck things in according to its conception of the world.
See also Coda Hale's metrics (https://github.com/codahale/metrics/).
-
HTTP via Goliath
-
also opens a UDP socket
-
is drop-in compatible with statsd
-
"Ruby Metrics":https://github.com/johnewart/ruby-metrics for accumulation
- ATTN: does this allow for namespacing?
-
drops packets but doesn't fail if Graphite is missing
-
drops packets but doesn't fail if MongoDB is missing
zeromq pubsub? pubsubhubbub?
- PUT /_stats/meter?name={name.of.stat}&incr={num}
- PUT /_stats/counter?name={name.of.stat}&incr={num}
- GET /_stats/instrument?name.of.stat=value
UDP HTTP facts metrics ruby eventmachine y y y y ruby stdlib-only y y ruby faraday n y y y java ? y y y curl (shell) n y y y
- Graphite
- MongoDB
Integrations can
-
send to a
-
store some limited
-
goliath: statsd_middleware and statsd_plugin.
-
ATTN: rack: ???
-
ATTN: rails: ???
QUESTIONS:
- ATTN: do we have to worry about memory use?
- ATTN: idempotency
- ATTN: make it easy to include, or ignore:
- machine info
- deploy info
The first path segment defines a collection, and the remainder defines a materialized path. All hashes within a given colxn.path should always have the same structure (so, don't record a george download and a george signup in the same scope).
For example a POST to http://vayacondios.whatever.com/code/commit with
{ "_id": "f93f2f08a0e39648fe64", # commit SHA as unique id
"_path": "code/commit", # materialized path
"_ts": "20110614104817", # utc flat time
"repo": "infochimps/wukong"
"message": "...",
"lines": 69,
"author": "mrflip", }
will write the hash as shown into the code
collection. Vayacondios fills in the _path always, and the _id and _ts if missing. This can be queried with path of "^commit/infochimps" or "^commit/.*".
Hash will hold:
-
_id
unique _id -
_ts
timestamp, set if blank -
_path
name.spaced.path.fact_name
, omits the collection part
- value
- count
- timing
- fact
echo 'bob.dobolina.mr.bob.dobolina:320|ms:320' | netcat -c -u 127.0.0.1 8125
http://bro:9000/bob/dobolina/mr/bob/dobolina?_count=1 http://bro:9000/bob/dobolina/mr/bob/dobolina?_count=1&_sampling=0.1 http://bro:9000/bob/dobolina/mr/bob/dobolina?_timing=320
-
path components must be ASCII strings matching
[a-z][a-z0-9_]+
-- that is, start with [a-z] and contain only lowercase alphanum or underscore. Components starting with a '' have reserved meanings -- the only _underscored fields that request can fill in are _id, _ts and _path. -
Vayacondios reserves the right to read and write paths in /bro, and the details of those paths will be documented; it will never read or write other paths unless explicitly asked to.
-
tree_merge rules:
-
hsh1 + hs## = hsh1.merge(hs##
-
arr1 + arr2 = arr1 + arr2
-
val1 + val2 = val2
-
hsh1 + nil = hsh1
-
arr1 + nil = arr1
-
val1 + nil = val1
-
nil + hs## = hsh2
-
nil + arr2 = arr2
-
nil + val2 = val2
-
otherwise, exception
types: Hash, Array, String, Time, Integer, Float, Nil
mongo: string, int, double, boolean, date, bytearray, object, array, others couch: string,number,boolean,array,object
-
h4. add (set? send?)
GET http://vayacondios:9099/f/{clxn}/{arbitrary/name/space}.{ext}?auth=token&query=predicate
db.collection(collection).save
get
h4. get
POST http://vayacondios:9099/f/arbitrary/name/space with JSON body
h4. increment
h4. add to set
what do we need to provide the 'I got next' feature (to distribute resources uniquely from a set)?
h4. auth
/_bro/_auth/
holds one hash giving public key.
- walk down the hash until you see _tok
- can only auth at first or second level?
- or by wildcard?
- access is read or read_write; by default allows read_write
GET latest GET all GET next
-
HTTP is connectionful:
- this is good: you get acknowledgement that a metric was recorded (this is good).
- this is bad: if the network is down, your code will break. (Well, usually. For some accounting and auditing metrics one might rather serve nothing at all than something unrecorded. Vayacondios doesn't address this use case.)
-
HTTP is ubiquitous, and allows for security, load-balancing and so forth
-
HTTP is fairly heavy-weight as RPC protocols go: you'll want to stick with UDP or streaming at the thousands-per-second case.
-
UDP is connectionless, so it's fire-and-forget. As Etsy puts it,
So, why do we use UDP to send data to StatsD? Well, it’s fast — you don’t want to slow your application down in order to track its performance — but also sending a UDP packet is fire-and-forget. Either StatsD gets the data, or it doesn’t. The application doesn’t care if StatsD is up, down, or on fire; it simply trusts that things will work. If they don’t, our stats go a bit wonky, but the site stays up. Because we also worship at the Church of Uptime, this is quite alright. (The Church of Graphs makes sure we graph UDP packet receipt failures though, which the kernel usefully provides.)
-
UDP, though bone-simple, is somewhat arcane
-
UDP has Packet Size limitations -- don't send large Events via UDP. From the EventMachine docs:
You may not send an arbitrarily-large data packet because your operating system will enforce a platform-specific limit on the size of the outbound packet. (Your kernel will respond in a platform-specific way if you send an overlarge packet: some will send a truncated packet, some will complain, and some will silently drop your request). On LANs, it’s usually OK to send datagrams up to about 4000 bytes in length, but to be really safe, send messages smaller than the Ethernet-packet size (typically about 1400 bytes). Some very restrictive WANs will either drop or truncate packets larger than about 500 bytes.
- Check out the latest master to make sure the feature hasn't been implemented or the bug hasn't been fixed yet
- Check out the issue tracker to make sure someone already hasn't requested it and/or contributed it
- Fork the project
- Start a feature/bugfix branch
- Commit and push until you are happy with your contribution
- Make sure to add tests for it. This is important so I don't break it in a future version unintentionally.
- Please try not to mess with the Rakefile, version, or history. If you want to have your own version, or is otherwise necessary, that is fine, but please isolate to its own commit so I can cherry-pick around it.
Copyright (c) 2011 Infochimps. See [LICENSE.md] for further details.