validation.go

Documentation: internal/trace/v2/testtrace

     1  // Copyright 2023 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package testtrace
     6  
     7  import (
     8  	"errors"
     9  	"fmt"
    10  	"internal/trace/v2"
    11  	"slices"
    12  	"strings"
    13  )
    14  
    15  // Validator is a type used for validating a stream of trace.Events.
    16  type Validator struct {
    17  	lastTs   trace.Time
    18  	gs       map[trace.GoID]*goState
    19  	ps       map[trace.ProcID]*procState
    20  	ms       map[trace.ThreadID]*schedContext
    21  	ranges   map[trace.ResourceID][]string
    22  	tasks    map[trace.TaskID]string
    23  	seenSync bool
    24  }
    25  
    26  type schedContext struct {
    27  	M trace.ThreadID
    28  	P trace.ProcID
    29  	G trace.GoID
    30  }
    31  
    32  type goState struct {
    33  	state   trace.GoState
    34  	binding *schedContext
    35  }
    36  
    37  type procState struct {
    38  	state   trace.ProcState
    39  	binding *schedContext
    40  }
    41  
    42  // NewValidator creates a new Validator.
    43  func NewValidator() *Validator {
    44  	return &Validator{
    45  		gs:     make(map[trace.GoID]*goState),
    46  		ps:     make(map[trace.ProcID]*procState),
    47  		ms:     make(map[trace.ThreadID]*schedContext),
    48  		ranges: make(map[trace.ResourceID][]string),
    49  		tasks:  make(map[trace.TaskID]string),
    50  	}
    51  }
    52  
    53  // Event validates ev as the next event in a stream of trace.Events.
    54  //
    55  // Returns an error if validation fails.
    56  func (v *Validator) Event(ev trace.Event) error {
    57  	e := new(errAccumulator)
    58  
    59  	// Validate timestamp order.
    60  	if v.lastTs != 0 {
    61  		if ev.Time() <= v.lastTs {
    62  			e.Errorf("timestamp out-of-order for %+v", ev)
    63  		} else {
    64  			v.lastTs = ev.Time()
    65  		}
    66  	} else {
    67  		v.lastTs = ev.Time()
    68  	}
    69  
    70  	// Validate event stack.
    71  	checkStack(e, ev.Stack())
    72  
    73  	switch ev.Kind() {
    74  	case trace.EventSync:
    75  		// Just record that we've seen a Sync at some point.
    76  		v.seenSync = true
    77  	case trace.EventMetric:
    78  		m := ev.Metric()
    79  		if !strings.Contains(m.Name, ":") {
    80  			// Should have a ":" as per runtime/metrics convention.
    81  			e.Errorf("invalid metric name %q", m.Name)
    82  		}
    83  		// Make sure the value is OK.
    84  		if m.Value.Kind() == trace.ValueBad {
    85  			e.Errorf("invalid value")
    86  		}
    87  		switch m.Value.Kind() {
    88  		case trace.ValueUint64:
    89  			// Just make sure it doesn't panic.
    90  			_ = m.Value.Uint64()
    91  		}
    92  	case trace.EventLabel:
    93  		l := ev.Label()
    94  
    95  		// Check label.
    96  		if l.Label == "" {
    97  			e.Errorf("invalid label %q", l.Label)
    98  		}
    99  
   100  		// Check label resource.
   101  		if l.Resource.Kind == trace.ResourceNone {
   102  			e.Errorf("label resource none")
   103  		}
   104  		switch l.Resource.Kind {
   105  		case trace.ResourceGoroutine:
   106  			id := l.Resource.Goroutine()
   107  			if _, ok := v.gs[id]; !ok {
   108  				e.Errorf("label for invalid goroutine %d", id)
   109  			}
   110  		case trace.ResourceProc:
   111  			id := l.Resource.Proc()
   112  			if _, ok := v.ps[id]; !ok {
   113  				e.Errorf("label for invalid proc %d", id)
   114  			}
   115  		case trace.ResourceThread:
   116  			id := l.Resource.Thread()
   117  			if _, ok := v.ms[id]; !ok {
   118  				e.Errorf("label for invalid thread %d", id)
   119  			}
   120  		}
   121  	case trace.EventStackSample:
   122  		// Not much to check here. It's basically a sched context and a stack.
   123  		// The sched context is also not guaranteed to align with other events.
   124  		// We already checked the stack above.
   125  	case trace.EventStateTransition:
   126  		// Validate state transitions.
   127  		//
   128  		// TODO(mknyszek): A lot of logic is duplicated between goroutines and procs.
   129  		// The two are intentionally handled identically; from the perspective of the
   130  		// API, resources all have the same general properties. Consider making this
   131  		// code generic over resources and implementing validation just once.
   132  		tr := ev.StateTransition()
   133  		checkStack(e, tr.Stack)
   134  		switch tr.Resource.Kind {
   135  		case trace.ResourceGoroutine:
   136  			// Basic state transition validation.
   137  			id := tr.Resource.Goroutine()
   138  			old, new := tr.Goroutine()
   139  			if new == trace.GoUndetermined {
   140  				e.Errorf("transition to undetermined state for goroutine %d", id)
   141  			}
   142  			if v.seenSync && old == trace.GoUndetermined {
   143  				e.Errorf("undetermined goroutine %d after first global sync", id)
   144  			}
   145  			if new == trace.GoNotExist && v.hasAnyRange(trace.MakeResourceID(id)) {
   146  				e.Errorf("goroutine %d died with active ranges", id)
   147  			}
   148  			state, ok := v.gs[id]
   149  			if ok {
   150  				if old != state.state {
   151  					e.Errorf("bad old state for goroutine %d: got %s, want %s", id, old, state.state)
   152  				}
   153  				state.state = new
   154  			} else {
   155  				if old != trace.GoUndetermined && old != trace.GoNotExist {
   156  					e.Errorf("bad old state for unregistered goroutine %d: %s", id, old)
   157  				}
   158  				state = &goState{state: new}
   159  				v.gs[id] = state
   160  			}
   161  			// Validate sched context.
   162  			if new.Executing() {
   163  				ctx := v.getOrCreateThread(e, ev.Thread())
   164  				if ctx != nil {
   165  					if ctx.G != trace.NoGoroutine && ctx.G != id {
   166  						e.Errorf("tried to run goroutine %d when one was already executing (%d) on thread %d", id, ctx.G, ev.Thread())
   167  					}
   168  					ctx.G = id
   169  					state.binding = ctx
   170  				}
   171  			} else if old.Executing() && !new.Executing() {
   172  				if tr.Stack != ev.Stack() {
   173  					// This is a case where the transition is happening to a goroutine that is also executing, so
   174  					// these two stacks should always match.
   175  					e.Errorf("StateTransition.Stack doesn't match Event.Stack")
   176  				}
   177  				ctx := state.binding
   178  				if ctx != nil {
   179  					if ctx.G != id {
   180  						e.Errorf("tried to stop goroutine %d when it wasn't currently executing (currently executing %d) on thread %d", id, ctx.G, ev.Thread())
   181  					}
   182  					ctx.G = trace.NoGoroutine
   183  					state.binding = nil
   184  				} else {
   185  					e.Errorf("stopping goroutine %d not bound to any active context", id)
   186  				}
   187  			}
   188  		case trace.ResourceProc:
   189  			// Basic state transition validation.
   190  			id := tr.Resource.Proc()
   191  			old, new := tr.Proc()
   192  			if new == trace.ProcUndetermined {
   193  				e.Errorf("transition to undetermined state for proc %d", id)
   194  			}
   195  			if v.seenSync && old == trace.ProcUndetermined {
   196  				e.Errorf("undetermined proc %d after first global sync", id)
   197  			}
   198  			if new == trace.ProcNotExist && v.hasAnyRange(trace.MakeResourceID(id)) {
   199  				e.Errorf("proc %d died with active ranges", id)
   200  			}
   201  			state, ok := v.ps[id]
   202  			if ok {
   203  				if old != state.state {
   204  					e.Errorf("bad old state for proc %d: got %s, want %s", id, old, state.state)
   205  				}
   206  				state.state = new
   207  			} else {
   208  				if old != trace.ProcUndetermined && old != trace.ProcNotExist {
   209  					e.Errorf("bad old state for unregistered proc %d: %s", id, old)
   210  				}
   211  				state = &procState{state: new}
   212  				v.ps[id] = state
   213  			}
   214  			// Validate sched context.
   215  			if new.Executing() {
   216  				ctx := v.getOrCreateThread(e, ev.Thread())
   217  				if ctx != nil {
   218  					if ctx.P != trace.NoProc && ctx.P != id {
   219  						e.Errorf("tried to run proc %d when one was already executing (%d) on thread %d", id, ctx.P, ev.Thread())
   220  					}
   221  					ctx.P = id
   222  					state.binding = ctx
   223  				}
   224  			} else if old.Executing() && !new.Executing() {
   225  				ctx := state.binding
   226  				if ctx != nil {
   227  					if ctx.P != id {
   228  						e.Errorf("tried to stop proc %d when it wasn't currently executing (currently executing %d) on thread %d", id, ctx.P, ctx.M)
   229  					}
   230  					ctx.P = trace.NoProc
   231  					state.binding = nil
   232  				} else {
   233  					e.Errorf("stopping proc %d not bound to any active context", id)
   234  				}
   235  			}
   236  		}
   237  	case trace.EventRangeBegin, trace.EventRangeActive, trace.EventRangeEnd:
   238  		// Validate ranges.
   239  		r := ev.Range()
   240  		switch ev.Kind() {
   241  		case trace.EventRangeBegin:
   242  			if v.hasRange(r.Scope, r.Name) {
   243  				e.Errorf("already active range %q on %v begun again", r.Name, r.Scope)
   244  			}
   245  			v.addRange(r.Scope, r.Name)
   246  		case trace.EventRangeActive:
   247  			if !v.hasRange(r.Scope, r.Name) {
   248  				v.addRange(r.Scope, r.Name)
   249  			}
   250  		case trace.EventRangeEnd:
   251  			if !v.hasRange(r.Scope, r.Name) {
   252  				e.Errorf("inactive range %q on %v ended", r.Name, r.Scope)
   253  			}
   254  			v.deleteRange(r.Scope, r.Name)
   255  		}
   256  	case trace.EventTaskBegin:
   257  		// Validate task begin.
   258  		t := ev.Task()
   259  		if t.ID == trace.NoTask || t.ID == trace.BackgroundTask {
   260  			// The background task should never have an event emitted for it.
   261  			e.Errorf("found invalid task ID for task of type %s", t.Type)
   262  		}
   263  		if t.Parent == trace.BackgroundTask {
   264  			// It's not possible for a task to be a subtask of the background task.
   265  			e.Errorf("found background task as the parent for task of type %s", t.Type)
   266  		}
   267  		// N.B. Don't check the task type. Empty string is a valid task type.
   268  		v.tasks[t.ID] = t.Type
   269  	case trace.EventTaskEnd:
   270  		// Validate task end.
   271  		// We can see a task end without a begin, so ignore a task without information.
   272  		// Instead, if we've seen the task begin, just make sure the task end lines up.
   273  		t := ev.Task()
   274  		if typ, ok := v.tasks[t.ID]; ok {
   275  			if t.Type != typ {
   276  				e.Errorf("task end type %q doesn't match task start type %q for task %d", t.Type, typ, t.ID)
   277  			}
   278  			delete(v.tasks, t.ID)
   279  		}
   280  	case trace.EventLog:
   281  		// There's really not much here to check, except that we can
   282  		// generate a Log. The category and message are entirely user-created,
   283  		// so we can't make any assumptions as to what they are. We also
   284  		// can't validate the task, because proving the task's existence is very
   285  		// much best-effort.
   286  		_ = ev.Log()
   287  	}
   288  	return e.Errors()
   289  }
   290  
   291  func (v *Validator) hasRange(r trace.ResourceID, name string) bool {
   292  	ranges, ok := v.ranges[r]
   293  	return ok && slices.Contains(ranges, name)
   294  }
   295  
   296  func (v *Validator) addRange(r trace.ResourceID, name string) {
   297  	ranges, _ := v.ranges[r]
   298  	ranges = append(ranges, name)
   299  	v.ranges[r] = ranges
   300  }
   301  
   302  func (v *Validator) hasAnyRange(r trace.ResourceID) bool {
   303  	ranges, ok := v.ranges[r]
   304  	return ok && len(ranges) != 0
   305  }
   306  
   307  func (v *Validator) deleteRange(r trace.ResourceID, name string) {
   308  	ranges, ok := v.ranges[r]
   309  	if !ok {
   310  		return
   311  	}
   312  	i := slices.Index(ranges, name)
   313  	if i < 0 {
   314  		return
   315  	}
   316  	v.ranges[r] = slices.Delete(ranges, i, i+1)
   317  }
   318  
   319  func (v *Validator) getOrCreateThread(e *errAccumulator, m trace.ThreadID) *schedContext {
   320  	if m == trace.NoThread {
   321  		e.Errorf("must have thread, but thread ID is none")
   322  		return nil
   323  	}
   324  	s, ok := v.ms[m]
   325  	if !ok {
   326  		s = &schedContext{M: m, P: trace.NoProc, G: trace.NoGoroutine}
   327  		v.ms[m] = s
   328  		return s
   329  	}
   330  	return s
   331  }
   332  
   333  func checkStack(e *errAccumulator, stk trace.Stack) {
   334  	// Check for non-empty values, but we also check for crashes due to incorrect validation.
   335  	i := 0
   336  	stk.Frames(func(f trace.StackFrame) bool {
   337  		if i == 0 {
   338  			// Allow for one fully zero stack.
   339  			//
   340  			// TODO(mknyszek): Investigate why that happens.
   341  			return true
   342  		}
   343  		if f.Func == "" || f.File == "" || f.PC == 0 || f.Line == 0 {
   344  			e.Errorf("invalid stack frame %#v: missing information", f)
   345  		}
   346  		i++
   347  		return true
   348  	})
   349  }
   350  
   351  type errAccumulator struct {
   352  	errs []error
   353  }
   354  
   355  func (e *errAccumulator) Errorf(f string, args ...any) {
   356  	e.errs = append(e.errs, fmt.Errorf(f, args...))
   357  }
   358  
   359  func (e *errAccumulator) Errors() error {
   360  	return errors.Join(e.errs...)
   361  }
   362
View as plain text