Tarantool tx thread event loop speed

[filonenko-mikhail]

Motivation

In case when there is some heavy fiber or fibers amount is big it's possible that fiber takes control too late.

For e.g. if fiber has some timeout logic, timeout can be triggered even if necessary data is arrived in buffer.

for e.g.

while clock.now() < deadline do
  local payload = sock.read(100, timeout=0.1)
end
-- payload is ignored
if clock.now() >= deadline then
  error
end

Propolsal

Measure tx thread event loop speed with special fiber

require('strict')
local log = require('log')
local clock = require('clock')
local fiber = require('fiber')
local metrics = require('metrics')

local timeout = 1

local tx_loop_delay_gauge = metrics.gauge('tnt_tx_loop_delay', 'Tarantool tx thread event loop delay')
if rawget(_G, '__metrics_tx_speed') == nil then
    _G.__metrics_tx_speed = fiber.new(function ()
            fiber.self():name("tx_speed")

            while true do
                local start = clock.monotonic64()
                fiber.sleep(tonumber(timeout))
                local diff = clock.monotonic64() - start
                local err = (tonumber(diff) - timeout*1e9)/1e9

                if err < 0 then
                    --log.info("Early wakeup %f", err)
                elseif err > 0 then
                    --log.info('Event loop slower by %f seconds', err)
                end
                tx_loop_delay_gauge:set(timeout + err)
            end
    end)
end

[github-actions]

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 30 days

[github-actions]

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 30 days