Go concurrency practice

Go Concurrency Practice Rules

Context & Goroutine Lifecycle

#61: Do not propagate a cancel-bound context into a goroutine that must outlive it

When spawning a goroutine from an HTTP handler (or any scoped context) to do async work (e.g., publish to Kafka, send metrics), do not pass the request context directly. The context cancels when the response is written, racing with the async work.

Rules:

• If the goroutine must outlive the parent scope, use context.Background() or a detached context that preserves values but strips cancellation.
• If values (trace IDs, correlation IDs) must be carried, create a detach wrapper:

// detach strips cancellation/deadline but keeps values.
type detach struct{ ctx context.Context }

func (d detach) Deadline() (time.Time, bool) { return time.Time{}, false }
func (d detach) Done() <-chan struct{}        { return nil }
func (d detach) Err() error                  { return nil }
func (d detach) Value(key any) any           { return d.ctx.Value(key) }

// BAD - context cancels when response is written
go func() { publish(r.Context(), response) }()

// GOOD - detach cancellation, keep values
go func() { publish(detach{ctx: r.Context()}, response) }()

#62: Every goroutine must have a known stopping condition

Before writing go func(), answer: when does this goroutine stop?

Rules:

• If a goroutine holds resources (connections, files), the parent must wait for cleanup, not just signal. Use defer w.close() or a sync.WaitGroup.
• Signaling (context cancellation) is not enough on its own -- it does not guarantee the goroutine had time to release resources.
• Return the goroutine's owner (struct with a Close method) so the caller can manage its lifetime.

// BAD - no way to wait for cleanup
func newWatcher(ctx context.Context) {
    w := watcher{}
    go w.watch(ctx)
}

// GOOD - caller controls shutdown
func newWatcher() watcher {
    w := watcher{}
    go w.watch()
    return w
}
// in main:
w := newWatcher()
defer w.close()

Goroutines & Loop Variables

#63: Capture loop variables before launching goroutines

Closure goroutines capture the variable, not the value. All goroutines share the same loop variable and will read whatever value it holds at execution time.

Rules:

• Create a local copy: val := i before the go func().
• Or pass the variable as a function argument: go func(v int) { ... }(i).

Note: Go 1.22+ changed for loop variable semantics so each iteration gets a new variable. If the module's go directive is >= 1.22, this is no longer a bug, but applying the fix is still safe and communicates intent.

// BAD (before Go 1.22)
for _, i := range s {
    go func() { fmt.Print(i) }() // may print same value repeatedly
}

// GOOD - local copy
for _, i := range s {
    val := i
    go func() { fmt.Print(val) }()
}

// GOOD - function argument
for _, i := range s {
    go func(val int) { fmt.Print(val) }(i)
}

Channel Patterns

#64: select on multiple channels picks randomly, not by source order

When multiple cases in a select can proceed, Go picks one uniformly at random. Writing case <-messageCh above case <-disconnectCh does not create priority.

Rules:

• For a single producer: use an unbuffered channel or a single channel carrying both message types to enforce ordering.
• For multiple producers needing priority drain: after receiving the low-priority signal, use an inner for/select with default to drain the high-priority channel.

// Priority drain pattern
for {
    select {
    case v := <-messageCh:
        fmt.Println(v)
    case <-disconnectCh:
        // Drain remaining messages before returning
        for {
            select {
            case v := <-messageCh:
                fmt.Println(v)
            default:
                return
            }
        }
    }
}

#65: Use chan struct{} for notification channels

If a channel carries no meaningful data (only the event of receiving matters), use chan struct{}, not chan bool. An empty struct is zero bytes; bool wastes space and creates ambiguity about what false means.

#66: Use nil channels to disable select cases

Receiving from or sending to a nil channel blocks forever. This is useful: after a channel is closed, set it to nil to remove that case from a select.

Key pattern -- merging channels:

func merge(ch1, ch2 <-chan int) <-chan int {
    ch := make(chan int, 1)
    go func() {
        for ch1 != nil || ch2 != nil {
            select {
            case v, open := <-ch1:
                if !open {
                    ch1 = nil // disable this case
                    break
                }
                ch <- v
            case v, open := <-ch2:
                if !open {
                    ch2 = nil
                    break
                }
                ch <- v
            }
        }
        close(ch)
    }()
    return ch
}

Without nil channels the closed-channel case fires continuously in a busy loop (receiving the zero value), wasting CPU.

#67: Default buffered channel size should be 1

• Use unbuffered channels when you need synchronization (sender blocks until receiver is ready) or for close-based notifications.
• Use buffered with size 1 as the default when a buffer is needed.
• Only use larger sizes for: worker pools (tie to goroutine count) or rate limiting (tie to the limit). Document the rationale for any magic number.
• Unbuffered channels are easier to reason about; buffered channels can hide deadlocks.

String Formatting Side Effects

#68: fmt functions can cause data races and deadlocks in concurrent code

Data race (etcd pattern): fmt.Sprintf("%v", ctx) traverses all values in a context chain. If any value is a mutable pointer and another goroutine mutates it concurrently, this is a data race. Avoid formatting entire contexts; extract only the specific immutable key you need.

Deadlock (Stringer + mutex pattern): If a method holds a mutex lock and formats %v on the receiver, and the receiver's String() method also acquires the same lock, you get a deadlock.

// BAD - deadlock: UpdateAge holds Lock, fmt calls String which calls RLock
func (c *Customer) UpdateAge(age int) error {
    c.mutex.Lock()
    defer c.mutex.Unlock()
    if age < 0 {
        return fmt.Errorf("age should be positive for customer %v", c)
        //                                                       ^^ calls String()
    }
    c.age = age
    return nil
}

func (c *Customer) String() string {
    c.mutex.RLock() // DEADLOCK: Lock already held
    defer c.mutex.RUnlock()
    return fmt.Sprintf("id %s, age %d", c.id, c.age)
}

Fixes:

1. Validate inputs before acquiring the lock.
2. In the error message, access fields directly (c.id) instead of formatting the whole struct.
3. Never call a method that acquires the same lock from within a locked section.

Data Races with Slices and Maps

#69: append is not always data-race-free

append on a full slice (len == cap) allocates a new backing array -- race-free. append on a non-full slice mutates the existing backing array -- data race if concurrent.

Rules:

• Never use append concurrently on a shared slice unless you are certain the slice is full (and you almost never are).
• If multiple goroutines need to append, give each a copy of the slice.

// BAD - data race when cap > len
s := make([]int, 0, 1)
go func() { s1 := append(s, 1); _ = s1 }()
go func() { s2 := append(s, 1); _ = s2 }()

// GOOD - each goroutine works on a copy
go func() {
    sCopy := make([]int, len(s), cap(s))
    copy(sCopy, s)
    s1 := append(sCopy, 1)
    _ = s1
}()

Map note: Any concurrent access to a map with at least one writer is a data race, regardless of key. Different slice indices are fine; different map keys are not.

#70: Assigning a slice or map to a local variable does not copy the data

balances := c.balances copies the header/pointer, not the underlying data. Both variables share the same backing storage. Mutating one is visible (and racy) through the other.

Rules:

• If the operation is lightweight, hold the lock for the entire operation.
• If the operation is heavy, deep-copy the data inside the lock, then release and operate on the copy.

// BAD - balances shares backing data with c.balances
func (c *Cache) AverageBalance() float64 {
    c.mu.RLock()
    balances := c.balances // NOT a deep copy
    c.mu.RUnlock()
    // iterating here races with writers
}

// GOOD option 1 - hold lock for entire read
func (c *Cache) AverageBalance() float64 {
    c.mu.RLock()
    defer c.mu.RUnlock()
    sum := 0.
    for _, b := range c.balances { sum += b }
    return sum / float64(len(c.balances))
}

// GOOD option 2 - deep copy then release
func (c *Cache) AverageBalance() float64 {
    c.mu.RLock()
    m := make(map[string]float64, len(c.balances))
    for k, v := range c.balances { m[k] = v }
    c.mu.RUnlock()
    // safe to iterate m without lock
}

Sync Primitives

#71: Call WaitGroup.Add before launching the goroutine

wg.Add(1) inside the goroutine is a race: wg.Wait() may return before the goroutine has even called Add.

Rules:

• Call wg.Add(n) in the parent goroutine, before go func().
• Call wg.Done() inside the goroutine (typically via defer).

// BAD
for i := 0; i < 3; i++ {
    go func() {
        wg.Add(1) // race: parent may call Wait() before this runs
        doWork()
        wg.Done()
    }()
}
wg.Wait()

// GOOD
for i := 0; i < 3; i++ {
    wg.Add(1)
    go func() {
        defer wg.Done()
        doWork()
    }()
}
wg.Wait()

#72: Use sync.Cond to broadcast to multiple waiting goroutines

Channels deliver each message to one goroutine (round-robin). Only a channel close is broadcast, and it can only happen once. When you need to repeatedly notify multiple goroutines of state changes, use sync.Cond.

Key API:

• sync.NewCond(&sync.Mutex{}) -- create with a Locker.
• cond.Wait() -- must be called inside cond.L.Lock()/Unlock(). Internally it unlocks, suspends, then re-locks on wake.
• cond.Broadcast() -- wakes all waiting goroutines.
• cond.Signal() -- wakes one waiting goroutine.

Caveat: If no goroutine is waiting when Broadcast/Signal is called, the notification is lost (unlike a buffered channel).

donation := &Donation{cond: sync.NewCond(&sync.Mutex{})}

// Listener
go func() {
    donation.cond.L.Lock()
    for donation.balance < goal {
        donation.cond.Wait()
    }
    fmt.Printf("$%d goal reached\n", donation.balance)
    donation.cond.L.Unlock()
}()

// Updater
donation.cond.L.Lock()
donation.balance++
donation.cond.L.Unlock()
donation.cond.Broadcast()

#73: Use errgroup for parallel goroutines that return errors

Do not hand-roll sync.WaitGroup + shared error slice + mutex for parallel work that can fail. Use golang.org/x/sync/errgroup.

Benefits:

• g.Go(func() error) -- spawns goroutine, collects first non-nil error.
• g.Wait() -- blocks until all goroutines finish, returns first error.
• errgroup.WithContext(ctx) -- creates a shared context canceled on first error, so remaining goroutines can bail out early. The goroutines must be context-aware for this to be effective.

g, ctx := errgroup.WithContext(ctx)
for i, item := range items {
    i, item := i, item
    g.Go(func() error {
        result, err := process(ctx, item)
        if err != nil {
            return err
        }
        results[i] = result
        return nil
    })
}
if err := g.Wait(); err != nil {
    return nil, err
}

#74: Never copy a sync type

All sync types must not be copied. This includes: sync.Mutex, sync.RWMutex, sync.WaitGroup, sync.Cond, sync.Map, sync.Once, sync.Pool.

Copying happens silently when:

1. A method has a value receiver on a struct containing a sync field.
2. A function takes a struct containing a sync field by value.
3. Assigning a struct containing a sync field to another variable.

Fixes:

• Use a pointer receiver on any method of a struct with sync fields.
• Or make the sync field a pointer (mu *sync.Mutex), but then you must initialize it explicitly in the constructor.

// BAD - value receiver copies the mutex
func (c Counter) Increment(name string) {
    c.mu.Lock()
    defer c.mu.Unlock()
    c.counters[name]++
}

// GOOD - pointer receiver
func (c *Counter) Increment(name string) {
    c.mu.Lock()
    defer c.mu.Unlock()
    c.counters[name]++
}

Run go vet or enable the copylocks analyzer -- it detects most sync copy issues at compile time.