Software /
code /
prosody
File
util/async.lua @ 11523:5f15ab7c6ae5
Statistics: Rewrite statistics backends to use OpenMetrics
The metric subsystem of Prosody has had some shortcomings from
the perspective of the current state-of-the-art in metric
observability.
The OpenMetrics standard [0] is a formalization of the data
model (and serialization format) of the well-known and
widely-used Prometheus [1] software stack.
The previous stats subsystem of Prosody did not map well to that
format (see e.g. [2] and [3]); the key reason is that it was
trying to do too much math on its own ([2]) while lacking
first-class support for "families" of metrics ([3]) and
structured metric metadata (despite the `extra` argument to
metrics, there was no standard way of representing common things
like "tags" or "labels").
Even though OpenMetrics has grown from the Prometheus world of
monitoring, it maps well to other popular monitoring stacks
such as:
- InfluxDB (labels can be mapped to tags and fields as necessary)
- Carbon/Graphite (labels can be attached to the metric name with
dot-separation)
- StatsD (see graphite when assuming that graphite is used as
backend, which is the default)
The util.statsd module has been ported to use the OpenMetrics
model as a proof of concept. An implementation which exposes
the util.statistics backend data as Prometheus metrics is
ready for publishing in prosody-modules (most likely as
mod_openmetrics_prometheus to avoid breaking existing 0.11
deployments).
At the same time, the previous measure()-based API had one major
advantage: It is really simple and easy to use without requiring
lots of knowledge about OpenMetrics or similar concepts. For that
reason as well as compatibility with existing code, it is preserved
and may even be extended in the future.
However, code relying on the `stats-updated` event as well as
`get_stats` from `statsmanager` will break because the data
model has changed completely; in case of `stats-updated`, the
code will simply not run (as the event was renamed in order
to avoid conflicts); the `get_stats` function has been removed
completely (so it will cause a traceback when it is attempted
to be used).
Note that the measure_*_event methods have been removed from
the module API. I was unable to find any uses or documentation
and thus deemed they should not be ported. Re-implementation is
possible when necessary.
[0]: https://openmetrics.io/
[1]: https://prometheus.io/
[2]: #959
[3]: #960
author | Jonas Schäfer <jonas@wielicki.name> |
---|---|
date | Sun, 18 Apr 2021 11:47:41 +0200 |
parent | 10931:558f0555ba02 |
child | 11961:542a9a503073 |
line wrap: on
line source
local logger = require "util.logger"; local log = logger.init("util.async"); local new_id = require "util.id".short; local xpcall = require "util.xpcall".xpcall; local function checkthread() local thread, main = coroutine.running(); if not thread or main then error("Not running in an async context, see https://prosody.im/doc/developers/util/async"); end return thread; end local function runner_from_thread(thread) local level = 0; -- Find the 'level' of the top-most function (0 == current level, 1 == caller, ...) while debug.getinfo(thread, level, "") do level = level + 1; end local name, runner = debug.getlocal(thread, level-1, 1); if name ~= "self" or type(runner) ~= "table" or runner.thread ~= thread then return nil; end return runner; end local function call_watcher(runner, watcher_name, ...) local watcher = runner.watchers[watcher_name]; if not watcher then return false; end runner:log("debug", "Calling '%s' watcher", watcher_name); local ok, err = xpcall(watcher, debug.traceback, runner, ...); if not ok then runner:log("error", "Error in '%s' watcher: %s", watcher_name, err); return nil, err; end return true; end local function runner_continue(thread) -- ASSUMPTION: runner is in 'waiting' state (but we don't have the runner to know for sure) if coroutine.status(thread) ~= "suspended" then -- This should suffice log("error", "unexpected async state: thread not suspended"); return false; end local ok, state, runner = coroutine.resume(thread); if not ok then local err = state; -- Running the coroutine failed, which means we have to find the runner manually, -- in order to inform the error handler runner = runner_from_thread(thread); if not runner then log("error", "unexpected async state: unable to locate runner during error handling"); return false; end call_watcher(runner, "error", debug.traceback(thread, err)); runner.state = "ready"; return runner:run(); elseif state == "ready" then -- If state is 'ready', it is our responsibility to update runner.state from 'waiting'. -- We also have to :run(), because the queue might have further items that will not be -- processed otherwise. FIXME: It's probably best to do this in a nexttick (0 timer). runner.state = "ready"; runner:run(); end return true; end local function waiter(num) local thread = checkthread(); num = num or 1; local waiting; return function () if num == 0 then return; end -- already done waiting = true; coroutine.yield("wait"); end, function () num = num - 1; if num == 0 and waiting then runner_continue(thread); elseif num < 0 then error("done() called too many times"); end end; end local function guarder() local guards = {}; local default_id = {}; return function (id, func) id = id or default_id; local thread = checkthread(); local guard = guards[id]; if not guard then guard = {}; guards[id] = guard; log("debug", "New guard!"); else table.insert(guard, thread); log("debug", "Guarded. %d threads waiting.", #guard) coroutine.yield("wait"); end local function exit() local next_waiting = table.remove(guard, 1); if next_waiting then log("debug", "guard: Executing next waiting thread (%d left)", #guard) runner_continue(next_waiting); else log("debug", "Guard off duty.") guards[id] = nil; end end if func then func(); exit(); return; end return exit; end; end local runner_mt = {}; runner_mt.__index = runner_mt; local function runner_create_thread(func, self) local thread = coroutine.create(function (self) -- luacheck: ignore 432/self while true do func(coroutine.yield("ready", self)); end end); debug.sethook(thread, debug.gethook()); assert(coroutine.resume(thread, self)); -- Start it up, it will return instantly to wait for the first input return thread; end local function default_error_watcher(runner, err) runner:log("error", "Encountered error: %s", err); error(err); end local function default_func(f) f(); end local function runner(func, watchers, data) local id = new_id(); local _log = logger.init("runner" .. id); return setmetatable({ func = func or default_func, thread = false, state = "ready", notified_state = "ready", queue = {}, watchers = watchers or { error = default_error_watcher }, data = data, id = id, _log = _log; } , runner_mt); end -- Add a task item for the runner to process function runner_mt:run(input) if input ~= nil then table.insert(self.queue, input); --self:log("debug", "queued new work item, %d items queued", #self.queue); end if self.state ~= "ready" then -- The runner is busy. Indicate that the task item has been -- queued, and return information about the current runner state return true, self.state, #self.queue; end local q, thread = self.queue, self.thread; if not thread or coroutine.status(thread) == "dead" then --luacheck: ignore 143/coroutine if thread and coroutine.close then coroutine.close(thread); end self:log("debug", "creating new coroutine"); -- Create a new coroutine for this runner thread = runner_create_thread(self.func, self); self.thread = thread; end -- Process task item(s) while the queue is not empty, and we're not blocked local n, state, err = #q, self.state, nil; self.state = "running"; --self:log("debug", "running main loop"); while n > 0 and state == "ready" and not err do local consumed; -- Loop through queue items, and attempt to run them for i = 1,n do local queued_input = q[i]; local ok, new_state = coroutine.resume(thread, queued_input); if not ok then -- There was an error running the coroutine, save the error, mark runner as ready to begin again consumed, state, err = i, "ready", debug.traceback(thread, new_state); self.thread = nil; break; elseif new_state == "wait" then -- Runner is blocked on waiting for a task item to complete consumed, state = i, "waiting"; break; end end -- Loop ended - either queue empty because all tasks passed without blocking (consumed == nil) -- or runner is blocked/errored, and consumed will contain the number of tasks processed so far if not consumed then consumed = n; end -- Remove consumed items from the queue array if q[n+1] ~= nil then n = #q; end for i = 1, n do q[i] = q[consumed+i]; end n = #q; end -- Runner processed all items it can, so save current runner state self.state = state; if err or state ~= self.notified_state then self:log("debug", "changed state from %s to %s", self.notified_state, err and ("error ("..state..")") or state); if err then state = "error" else self.notified_state = state; end local handler = self.watchers[state]; if handler then handler(self, err); end end if n > 0 then return self:run(); end return true, state, n; end -- Add a task item to the queue without invoking the runner, even if it is idle function runner_mt:enqueue(input) table.insert(self.queue, input); self:log("debug", "queued new work item, %d items queued", #self.queue); return self; end function runner_mt:log(level, fmt, ...) return self._log(level, fmt, ...); end function runner_mt:onready(f) self.watchers.ready = f; return self; end function runner_mt:onwaiting(f) self.watchers.waiting = f; return self; end function runner_mt:onerror(f) self.watchers.error = f; return self; end local function ready() return pcall(checkthread); end local function wait_for(promise) local async_wait, async_done = waiter(); local ret, err = nil, nil; promise:next( function (r) ret = r; end, function (e) err = e; end) :finally(async_done); async_wait(); if ret then return ret; else return nil, err; end end return { ready = ready; waiter = waiter; guarder = guarder; runner = runner; wait = wait_for; -- COMPAT w/trunk pre-0.12 wait_for = wait_for; };