| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package jobruntime |
| 4 | |
| 5 | import ( |
| 6 | "sync" |
| 7 | "sync/atomic" |
| 8 | "time" |
| 9 | |
| 10 | "github.com/netdata/netdata/go/plugins/logger" |
| 11 | "github.com/netdata/netdata/go/plugins/plugin/framework/tickstate" |
| 12 | ) |
| 13 | |
| 14 | const ( |
| 15 | penaltyStep = 5 |
| 16 | maxPenalty = 600 |
| 17 | infTries = -1 |
| 18 | ) |
| 19 | |
| 20 | // stopController coordinates stop request and completion acknowledgement. |
| 21 | // It is safe for repeated Stop calls and pre-start stop requests. |
| 22 | type stopController struct { |
| 23 | stopCh chan struct{} |
| 24 | stoppedCh chan struct{} |
| 25 | |
| 26 | started atomic.Bool |
| 27 | |
| 28 | stopOnce sync.Once |
| 29 | stoppedOnce sync.Once |
| 30 | } |
| 31 | |
| 32 | func newStopController() stopController { |
| 33 | return stopController{ |
| 34 | stopCh: make(chan struct{}), |
| 35 | stoppedCh: make(chan struct{}), |
| 36 | } |
| 37 | } |
| 38 | |
| 39 | func (c *stopController) markStarted() { |
| 40 | c.started.Store(true) |
| 41 | } |
| 42 | |
| 43 | func (c *stopController) markStopped() { |
| 44 | c.stoppedOnce.Do(func() { close(c.stoppedCh) }) |
| 45 | } |
| 46 | |
| 47 | func (c *stopController) requestStop() { |
| 48 | c.stopOnce.Do(func() { close(c.stopCh) }) |
| 49 | } |
| 50 | |
| 51 | func (c *stopController) stopAndWait() { |
| 52 | c.requestStop() |
| 53 | if !c.started.Load() { |
| 54 | return |
| 55 | } |
| 56 | <-c.stoppedCh |
| 57 | } |
| 58 | |
| 59 | func retryAutoDetection(autoDetectEvery, autoDetectTries int) bool { |
| 60 | return autoDetectEvery > 0 && (autoDetectTries == infTries || autoDetectTries > 0) |
| 61 | } |
| 62 | |
| 63 | func disableAutoDetection(autoDetectEvery *int) { |
| 64 | *autoDetectEvery = 0 |
| 65 | } |
| 66 | |
| 67 | func consumeAutoDetectTry(autoDetectTries *int) { |
| 68 | if *autoDetectTries != infTries { |
| 69 | *autoDetectTries-- |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | func shouldCollectWithPenalty(clock, updateEvery, retries int) bool { |
| 74 | return clock%(updateEvery+penaltyFromRetries(retries, updateEvery)) == 0 |
| 75 | } |
| 76 | |
| 77 | func penaltyFromRetries(retries, updateEvery int) int { |
| 78 | v := retries / penaltyStep * penaltyStep * updateEvery / 2 |
| 79 | if v > maxPenalty { |
| 80 | return maxPenalty |
| 81 | } |
| 82 | return v |
| 83 | } |
| 84 | |
| 85 | func enqueueTickWithSkipLog( |
| 86 | tick chan int, |
| 87 | clock int, |
| 88 | functionOnly bool, |
| 89 | updateEvery int, |
| 90 | retries int, |
| 91 | skipTracker *tickstate.SkipTracker, |
| 92 | log *logger.Logger, |
| 93 | ) { |
| 94 | select { |
| 95 | case tick <- clock: |
| 96 | default: |
| 97 | if functionOnly || !shouldCollectWithPenalty(clock, updateEvery, retries) { |
| 98 | return |
| 99 | } |
| 100 | |
| 101 | skip := skipTracker.MarkSkipped() |
| 102 | if skip.RunStarted.IsZero() { |
| 103 | log.Infof("skipping data collection: waiting for first collection to start (interval %ds)", updateEvery) |
| 104 | return |
| 105 | } |
| 106 | if skip.Count >= 2 { |
| 107 | log.Warningf( |
| 108 | "skipping data collection: previous run is still in progress for %s (skipped %d times in a row, interval %ds)", |
| 109 | time.Since(skip.RunStarted), |
| 110 | skip.Count, |
| 111 | updateEvery, |
| 112 | ) |
| 113 | return |
| 114 | } |
| 115 | log.Infof("skipping data collection: previous run is still in progress for %s (interval %ds)", time.Since(skip.RunStarted), updateEvery) |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | func markRunStartWithResumeLog(skipTracker *tickstate.SkipTracker, log *logger.Logger) { |
| 120 | resume := skipTracker.MarkRunStart(time.Now()) |
| 121 | if resume.Skipped == 0 { |
| 122 | return |
| 123 | } |
| 124 | if resume.RunStopped.IsZero() || resume.RunStarted.IsZero() { |
| 125 | log.Infof("data collection resumed (skipped %d times)", resume.Skipped) |
| 126 | return |
| 127 | } |
| 128 | log.Infof("data collection resumed after %s (skipped %d times)", resume.RunStopped.Sub(resume.RunStarted), resume.Skipped) |
| 129 | } |