Skip to content

Commit

Permalink
Enhance HA "Taking over", "Handing over" logging
Browse files Browse the repository at this point in the history 
The reason for a switch in the HA roles was not always directly clear.
This change now introduces additional logging, indicating the reasoning
for either taking over or handing over the HA responsibility.

Next to additional logging messages, both the takeover and handover
channel are now transporting a string to communicate the reason instead
of an empty struct{}. By doing so, both the "Taking over" and "Handing
over" log messages are enriched with a reason.

This also required a change in the suppressed logging handling of the
HA.realize method, which got its logging enabled through the shouldLog
parameter. Now, there are both recurring events, which might be
suppressed, as well as state changing events, which should be logged.
Therefore, and because the logTicker's functionality was not clear to me
on first glance, I renamed it to routineLogTicker.

While dealing with the code, some function signature documentation were
added, to ease both mine as well as the understanding of future readers.

Closes #688.
  • Loading branch information
@oxzi
oxzi committed Mar 12, 2024
1 parent 0e9810c commit 95d2a80
Show file tree
Hide file tree
Showing 2 changed files with 62 additions and 36 deletions.
8 changes: 4 additions & 4 deletions cmd/icingadb/main.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -161,8 +161,8 @@ func run() int {
hactx, cancelHactx := context.WithCancel(ctx)
for hactx.Err() == nil {
select {
case <-ha.Takeover():
logger.Info("Taking over")
case takeoverReason := <-ha.Takeover():
logger.Infow("Taking over", zap.String("reason", takeoverReason))

go func() {
for hactx.Err() == nil {
Expand Down Expand Up @@ -323,8 +323,8 @@ func run() int {
}
}
}()
case <-ha.Handover():
logger.Warn("Handing over")
case handoverReason := <-ha.Handover():
logger.Warnw("Handing over", zap.String("reason", handoverReason))

cancelHactx()
case <-hactx.Done():
Expand Down
90 changes: 58 additions & 32 deletions pkg/icingadb/ha.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,8 +43,8 @@ type HA struct {
heartbeat *icingaredis.Heartbeat
logger *logging.Logger
responsible bool
handover chan struct{}
takeover chan struct{}
handover chan string
takeover chan string
done chan struct{}
errOnce sync.Once
errMu sync.Mutex
Expand All @@ -64,8 +64,8 @@ func NewHA(ctx context.Context, db *DB, heartbeat *icingaredis.Heartbeat, logger
db: db,
heartbeat: heartbeat,
logger: logger,
handover: make(chan struct{}),
takeover: make(chan struct{}),
handover: make(chan string),
takeover: make(chan string),
done: make(chan struct{}),
}

Expand Down Expand Up @@ -107,13 +107,13 @@ func (h *HA) Err() error {
return h.err
}

// Handover returns a channel with which handovers are signaled.
func (h *HA) Handover() chan struct{} {
// Handover returns a channel with which handovers and their reasons are signaled.
func (h *HA) Handover() chan string {
return h.handover
}

// Takeover returns a channel with which takeovers are signaled.
func (h *HA) Takeover() chan struct{} {
// Takeover returns a channel with which takeovers and their reasons are signaled.
func (h *HA) Takeover() chan string {
return h.takeover
}

Expand Down Expand Up @@ -141,9 +141,10 @@ func (h *HA) controller() {

oldInstancesRemoved := false

logTicker := time.NewTicker(time.Second * 60)
defer logTicker.Stop()
shouldLog := true
// Suppress recurring log messages in the realize method to be only logged this often.
routineLogTicker := time.NewTicker(time.Second * 60)
defer routineLogTicker.Stop()
shouldLogRoutineEvents := true

for {
select {
Expand All @@ -160,7 +161,7 @@ func (h *HA) controller() {
}
if tt.Before(now.Add(-1 * timeout)) {
h.logger.Errorw("Received heartbeat from the past", zap.Time("time", tt))
h.signalHandover()
h.signalHandover("received heartbeat from the past")
h.realizeLostHeartbeat()
continue
}
Expand Down Expand Up @@ -197,8 +198,8 @@ func (h *HA) controller() {
}

select {
case <-logTicker.C:
shouldLog = true
case <-routineLogTicker.C:
shouldLogRoutineEvents = true
default:
}

Expand All @@ -209,10 +210,10 @@ func (h *HA) controller() {
} else {
realizeCtx, cancelRealizeCtx = context.WithCancel(h.ctx)
}
err = h.realize(realizeCtx, s, t, envId, shouldLog)
err = h.realize(realizeCtx, s, t, envId, shouldLogRoutineEvents)
cancelRealizeCtx()
if errors.Is(err, context.DeadlineExceeded) {
h.signalHandover()
h.signalHandover("context deadline exceeded")
continue
}
if err != nil {
Expand All @@ -224,10 +225,10 @@ func (h *HA) controller() {
oldInstancesRemoved = true
}

shouldLog = false
shouldLogRoutineEvents = false
} else {
h.logger.Error("Lost heartbeat")
h.signalHandover()
h.signalHandover("lost heartbeat")
h.realizeLostHeartbeat()
}
case <-h.heartbeat.Done():
Expand All @@ -240,13 +241,25 @@ func (h *HA) controller() {
}
}

func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *types.UnixMilli, envId types.Binary, shouldLog bool) error {
var takeover, otherResponsible bool
// realize a HA cycle triggered by a heartbeat event.
//
// shouldLogRoutineEvents indicates if recurrent events should be logged.
func (h *HA) realize(
ctx context.Context,
s *icingaredisv1.IcingaStatus,
t *types.UnixMilli,
envId types.Binary,
shouldLogRoutineEvents bool,
) error {
var (
takeover string
otherResponsible bool
)

err := retry.WithBackoff(
ctx,
func(ctx context.Context) error {
takeover = false
takeover = ""
otherResponsible = false
isoLvl := sql.LevelSerializable
selectLock := ""
Expand Down Expand Up @@ -274,15 +287,25 @@ func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *type
switch errQuery {
case nil:
otherResponsible = true
if shouldLog {
if shouldLogRoutineEvents {
h.logger.Infow("Another instance is active",
zap.String("instance_id", instance.Id.String()),
zap.String("environment", envId.String()),
"heartbeat", instance.Heartbeat,
zap.Time("heartbeat", instance.Heartbeat.Time()),
zap.Duration("heartbeat_age", time.Since(instance.Heartbeat.Time())))
}
case sql.ErrNoRows:
takeover = true
fields := []any{
zap.String("instance_id", instance.Id.String()),
zap.String("environment", envId.String()),
zap.Duration("heartbeat_timeout", timeout)}

if !h.responsible {
h.logger.Infow("Preparing to take over HA as no instance is active", fields...)
} else if h.responsible && shouldLogRoutineEvents {
h.logger.Infow("Continuing being the active instance", fields...)
}
takeover = "no other instance is active"
default:
return internal.CantPerformQuery(errQuery, query)
}
Expand All @@ -297,7 +320,7 @@ func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *type
EnvironmentId: envId,
},
Heartbeat: *t,
Responsible: types.Bool{Bool: takeover || h.responsible, Valid: true},
Responsible: types.Bool{Bool: takeover != "" || h.responsible, Valid: true},
EndpointId: s.EndpointId,
Icinga2Version: s.Version,
Icinga2StartTime: s.ProgramStart,
Expand All @@ -314,7 +337,7 @@ func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *type
return internal.CantPerformQuery(err, stmt)
}

if takeover {
if takeover != "" {
stmt := h.db.Rebind("UPDATE icingadb_instance SET responsible = ? WHERE environment_id = ? AND id <> ?")
_, err := tx.ExecContext(ctx, stmt, "n", envId, h.instanceId)

Expand Down Expand Up @@ -348,14 +371,14 @@ func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *type
return err
}

if takeover {
if takeover != "" {
// Insert the environment after each heartbeat takeover if it does not already exist in the database
// as the environment may have changed, although this is likely to happen very rarely.
if err := h.insertEnvironment(); err != nil {
return errors.Wrap(err, "can't insert environment")
}

h.signalTakeover()
h.signalTakeover(takeover)
} else if otherResponsible {
if state, _ := h.state.Load(); !state.otherResponsible {
state.otherResponsible = true
Expand All @@ -366,6 +389,7 @@ func (h *HA) realize(ctx context.Context, s *icingaredisv1.IcingaStatus, t *type
return nil
}

// realizeLostHeartbeat updates "responsible = n" for this HA into the database.
func (h *HA) realizeLostHeartbeat() {
stmt := h.db.Rebind("UPDATE icingadb_instance SET responsible = ? WHERE id = ?")
if _, err := h.db.ExecContext(h.ctx, stmt, "n", h.instanceId); err != nil && !utils.IsContextCanceled(err) {
Expand Down Expand Up @@ -421,7 +445,8 @@ func (h *HA) removeOldInstances(s *icingaredisv1.IcingaStatus, envId types.Binar
}
}

func (h *HA) signalHandover() {
// signalHandover gives up HA.responsible and notifies the HA.Handover chan.
func (h *HA) signalHandover(reason string) {
if h.responsible {
h.state.Store(haState{
responsibleTsMilli: time.Now().UnixMilli(),
Expand All @@ -430,15 +455,16 @@ func (h *HA) signalHandover() {
})

select {
case h.handover <- struct{}{}:
case h.handover <- reason:
h.responsible = false
case <-h.ctx.Done():
// Noop
}
}
}

func (h *HA) signalTakeover() {
// signalTakeover claims HA.responsible and notifies the HA.Takeover chan.
func (h *HA) signalTakeover(reason string) {
if !h.responsible {
h.state.Store(haState{
responsibleTsMilli: time.Now().UnixMilli(),
Expand All @@ -447,7 +473,7 @@ func (h *HA) signalTakeover() {
})

select {
case h.takeover <- struct{}{}:
case h.takeover <- reason:
h.responsible = true
case <-h.ctx.Done():
// Noop
Expand Down

0 comments on commit 95d2a80

Please sign in to comment.