| 1 | package lease |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "sync" |
| 7 | "time" |
| 8 | |
| 9 | "github.com/benbjohnson/litestream" |
| 10 | lss3 "github.com/benbjohnson/litestream/s3" |
| 11 | ) |
| 12 | |
| 13 | var ( |
| 14 | ErrLost = errors.New("lease safety deadline elapsed") |
| 15 | ErrSpent = errors.New("lease handle is spent") |
| 16 | ErrBusy = errors.New("lease operation already running") |
| 17 | ) |
| 18 | |
| 19 | const terminationMargin = 5 * time.Second |
| 20 | |
| 21 | // RenewMargin is the margin Held.SafeUntil subtracts from the lease's |
| 22 | // expiry at production TTLs: SafeUntil is at most TTL-RenewMargin ahead of |
| 23 | // a fresh acquire. The moves staging guard reasons from it. |
| 24 | const RenewMargin = terminationMargin |
| 25 | |
| 26 | // Production uses the fixed margin. Short TTLs exist only through AcquireTTL |
| 27 | // (tests; production callers pass LeaseTTL or the moves staging term). |
| 28 | func marginForTTL(ttl time.Duration) time.Duration { |
| 29 | if ttl < 3*terminationMargin { |
| 30 | return ttl / 10 |
| 31 | } |
| 32 | return terminationMargin |
| 33 | } |
| 34 | |
| 35 | type renewer struct { |
| 36 | ctx context.Context |
| 37 | leaser *lss3.Leaser |
| 38 | mu sync.Mutex |
| 39 | lease *litestream.Lease |
| 40 | err error |
| 41 | until time.Time |
| 42 | margin time.Duration |
| 43 | stopped bool |
| 44 | cancel context.CancelFunc |
| 45 | done chan struct{} |
| 46 | watchDone chan struct{} |
| 47 | changed chan struct{} |
| 48 | lost chan struct{} |
| 49 | } |
| 50 | |
| 51 | func startRenewer(leaser *lss3.Leaser, lease *litestream.Lease, started time.Time, margin time.Duration) *renewer { |
| 52 | ctx, cancel := context.WithCancel(context.Background()) |
| 53 | rn := &renewer{ctx: ctx, leaser: leaser, lease: lease, until: started.Add(leaser.TTL - margin), margin: margin, cancel: cancel, done: make(chan struct{}), watchDone: make(chan struct{}), changed: make(chan struct{}, 1), lost: make(chan struct{})} |
| 54 | go rn.watch(ctx) |
| 55 | go rn.renew(ctx) |
| 56 | return rn |
| 57 | } |
| 58 | |
| 59 | func (rn *renewer) loseLocked(err error) { |
| 60 | if rn.err == nil { |
| 61 | rn.err = err |
| 62 | close(rn.lost) |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | // checkLocked latches loss permanently; an acknowledgement arriving after a |
| 67 | // blocked renewal cannot resurrect ownership, even before the timer runs. |
| 68 | func (rn *renewer) checkLocked() error { |
| 69 | if rn.err != nil { |
| 70 | return rn.err |
| 71 | } |
| 72 | if rn.stopped { |
| 73 | return ErrSpent |
| 74 | } |
| 75 | if !time.Now().Before(rn.until) { |
| 76 | rn.loseLocked(ErrLost) |
| 77 | } |
| 78 | return rn.err |
| 79 | } |
| 80 | func (rn *renewer) Err() error { |
| 81 | rn.mu.Lock() |
| 82 | defer rn.mu.Unlock() |
| 83 | return rn.checkLocked() |
| 84 | } |
| 85 | func (rn *renewer) SafeUntil() (time.Time, error) { |
| 86 | rn.mu.Lock() |
| 87 | defer rn.mu.Unlock() |
| 88 | if err := rn.checkLocked(); err != nil { |
| 89 | return time.Time{}, err |
| 90 | } |
| 91 | return rn.until, nil |
| 92 | } |
| 93 | func (rn *renewer) watch(ctx context.Context) { |
| 94 | defer close(rn.watchDone) |
| 95 | for { |
| 96 | until, err := rn.SafeUntil() |
| 97 | if err != nil { |
| 98 | rn.cancel() |
| 99 | return |
| 100 | } |
| 101 | timer := time.NewTimer(time.Until(until)) |
| 102 | select { |
| 103 | case <-ctx.Done(): |
| 104 | timer.Stop() |
| 105 | return |
| 106 | case <-rn.changed: |
| 107 | timer.Stop() |
| 108 | case <-timer.C: |
| 109 | } |
| 110 | } |
| 111 | } |
| 112 | func (rn *renewer) renew(ctx context.Context) { |
| 113 | defer close(rn.done) |
| 114 | every := rn.leaser.TTL / 3 |
| 115 | if every <= 0 { |
| 116 | every = time.Nanosecond |
| 117 | } |
| 118 | ticker := time.NewTicker(every) |
| 119 | defer ticker.Stop() |
| 120 | for { |
| 121 | select { |
| 122 | case <-ctx.Done(): |
| 123 | return |
| 124 | case <-ticker.C: |
| 125 | rn.mu.Lock() |
| 126 | err := rn.checkLocked() |
| 127 | cur := rn.lease |
| 128 | rn.mu.Unlock() |
| 129 | if err != nil { |
| 130 | rn.cancel() |
| 131 | return |
| 132 | } |
| 133 | started := time.Now() |
| 134 | renewed, err := rn.leaser.RenewLease(ctx, cur) |
| 135 | rn.mu.Lock() |
| 136 | health := rn.checkLocked() |
| 137 | // Retain the newest ETag solely for conditional cleanup, including a late |
| 138 | // success. It never repairs a latched loss or extends a stopped handle. |
| 139 | if err == nil { |
| 140 | rn.lease = renewed |
| 141 | } |
| 142 | if health == nil { |
| 143 | if err != nil { |
| 144 | rn.loseLocked(err) |
| 145 | } else { |
| 146 | rn.until = started.Add(rn.leaser.TTL - rn.margin) |
| 147 | rn.checkLocked() |
| 148 | } |
| 149 | } |
| 150 | failed := rn.err != nil || rn.stopped |
| 151 | rn.mu.Unlock() |
| 152 | if failed { |
| 153 | rn.cancel() |
| 154 | return |
| 155 | } |
| 156 | select { |
| 157 | case rn.changed <- struct{}{}: |
| 158 | default: |
| 159 | } |
| 160 | } |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | func (rn *renewer) stop() *litestream.Lease { |
| 165 | rn.mu.Lock() |
| 166 | rn.checkLocked() |
| 167 | rn.stopped = true |
| 168 | rn.mu.Unlock() |
| 169 | rn.cancel() |
| 170 | <-rn.done |
| 171 | <-rn.watchDone |
| 172 | rn.mu.Lock() |
| 173 | defer rn.mu.Unlock() |
| 174 | return rn.lease |
| 175 | } |
| 176 | |