From cc6179775512a539155a1588ecad794ab8152b27 Mon Sep 17 00:00:00 2001 From: Daniel Jaglowski Date: Thu, 10 Jun 2021 16:52:15 -0400 Subject: [PATCH] Revert "track_rotated (#168)" (#180) This reverts commit c46512e6b2230c977e49b8ac6c00c8b6fb02f45c. --- docs/operators/file_input.md | 7 +- operator/builtin/input/file/benchmark_test.go | 2 +- operator/builtin/input/file/config.go | 4 +- operator/builtin/input/file/file.go | 72 ++++++++----------- operator/builtin/input/file/file_test.go | 19 ++--- operator/builtin/input/file/reader.go | 14 ++-- operator/builtin/input/file/rotation_test.go | 36 ---------- operator/builtin/input/file/util_test.go | 8 +-- 8 files changed, 51 insertions(+), 111 deletions(-) diff --git a/docs/operators/file_input.md b/docs/operators/file_input.md index 64f84c07..7f15089d 100644 --- a/docs/operators/file_input.md +++ b/docs/operators/file_input.md @@ -19,7 +19,7 @@ The `file_input` operator reads logs from files. It will place the lines read in | `start_at` | `end` | At startup, where to start reading logs from the file. Options are `beginning` or `end` | | `fingerprint_size` | `1kb` | The number of bytes with which to identify a file. The first bytes in the file are used as the fingerprint. Decreasing this value at any point will cause existing fingerprints to forgotten, meaning that all files will be read from the beginning (one time). | | `max_log_size` | `1MiB` | The maximum size of a log entry to read before failing. Protects against reading large amounts of data into memory | -| `max_concurrent_files` | 1024 | The maximum number of log files from which logs will be read concurrently (minimum = 2). If the number of files matched in the `include` pattern exceeds half of this number, then files will be processed in batches. One batch will be processed per `poll_interval`. | +| `max_concurrent_files` | 1024 | The maximum number of log files from which logs will be read concurrently. If the number of files matched in the `include` pattern exceeds this number, then files will be processed in batches. One batch will be processed per `poll_interval`. | | `attributes` | {} | A map of `key: value` pairs to add to the entry's attributes | | `resource` | {} | A map of `key: value` pairs to add to the entry's resource | @@ -37,11 +37,6 @@ match either the beginning of a new log entry, or the end of a log entry. Also refer to [recombine](/docs/operators/recombine.md) operator for merging events with greater control. -### File rotation - -When files are rotated and its new names are no longer captured in `include` pattern (i.e. tailing symlink files), it could result in data loss. -To avoid the data loss, choose move/create rotation method and set `max_concurrent_files` higher than the twice of the number of files to tail. - ### Supported encodings | Key | Description diff --git a/operator/builtin/input/file/benchmark_test.go b/operator/builtin/input/file/benchmark_test.go index 0656487d..5df09e48 100644 --- a/operator/builtin/input/file/benchmark_test.go +++ b/operator/builtin/input/file/benchmark_test.go @@ -104,7 +104,7 @@ func BenchmarkFileInput(b *testing.B) { cfg.Include = []string{ "file*.log", } - cfg.MaxConcurrentFiles = 2 + cfg.MaxConcurrentFiles = 1 return cfg }, }, diff --git a/operator/builtin/input/file/config.go b/operator/builtin/input/file/config.go index 59227a39..27daaf16 100644 --- a/operator/builtin/input/file/config.go +++ b/operator/builtin/input/file/config.go @@ -98,8 +98,8 @@ func (c InputConfig) Build(context operator.BuildContext) ([]operator.Operator, return nil, fmt.Errorf("`max_log_size` must be positive") } - if c.MaxConcurrentFiles <= 1 { - return nil, fmt.Errorf("`max_concurrent_files` must be greater than 1") + if c.MaxConcurrentFiles <= 0 { + return nil, fmt.Errorf("`max_concurrent_files` must be positive") } if c.FingerprintSize == 0 { diff --git a/operator/builtin/input/file/file.go b/operator/builtin/input/file/file.go index 0056cd91..52688822 100644 --- a/operator/builtin/input/file/file.go +++ b/operator/builtin/input/file/file.go @@ -49,10 +49,8 @@ type InputOperator struct { persister operator.Persister - knownFiles []*Reader - queuedMatches []string - maxBatchFiles int - lastPollReaders []*Reader + knownFiles []*Reader + queuedMatches []string startAtBeginning bool @@ -87,12 +85,6 @@ func (f *InputOperator) Start(persister operator.Persister) error { func (f *InputOperator) Stop() error { f.cancel() f.wg.Wait() - for _, reader := range f.lastPollReaders { - reader.Close() - } - for _, reader := range f.knownFiles { - reader.Close() - } f.knownFiles = nil f.cancel = nil return nil @@ -121,10 +113,9 @@ func (f *InputOperator) startPoller(ctx context.Context) { // poll checks all the watched paths for new entries func (f *InputOperator) poll(ctx context.Context) { - f.maxBatchFiles = f.MaxConcurrentFiles / 2 var matches []string - if len(f.queuedMatches) > f.maxBatchFiles { - matches, f.queuedMatches = f.queuedMatches[:f.maxBatchFiles], f.queuedMatches[f.maxBatchFiles:] + if len(f.queuedMatches) > f.MaxConcurrentFiles { + matches, f.queuedMatches = f.queuedMatches[:f.MaxConcurrentFiles], f.queuedMatches[f.MaxConcurrentFiles:] } else { if len(f.queuedMatches) > 0 { matches, f.queuedMatches = f.queuedMatches, make([]string, 0) @@ -139,8 +130,8 @@ func (f *InputOperator) poll(ctx context.Context) { matches = getMatches(f.Include, f.Exclude) if f.firstCheck && len(matches) == 0 { f.Warnw("no files match the configured include patterns", "include", f.Include) - } else if len(matches) > f.maxBatchFiles { - matches, f.queuedMatches = matches[:f.maxBatchFiles], matches[f.maxBatchFiles:] + } else if len(matches) > f.MaxConcurrentFiles { + matches, f.queuedMatches = matches[:f.MaxConcurrentFiles], matches[f.MaxConcurrentFiles:] } } } @@ -148,27 +139,7 @@ func (f *InputOperator) poll(ctx context.Context) { readers := f.makeReaders(matches) f.firstCheck = false - // Detect files that have been rotated out of matching pattern - lostReaders := make([]*Reader, 0, len(f.lastPollReaders)) -OUTER: - for _, oldReader := range f.lastPollReaders { - for _, reader := range readers { - if reader.Fingerprint.StartsWith(oldReader.Fingerprint) { - continue OUTER - } - } - lostReaders = append(lostReaders, oldReader) - } - var wg sync.WaitGroup - for _, reader := range lostReaders { - wg.Add(1) - go func(r *Reader) { - defer wg.Done() - r.ReadToEnd(ctx) - }(reader) - } - for _, reader := range readers { wg.Add(1) go func(r *Reader) { @@ -180,13 +151,6 @@ OUTER: // Wait until all the reader goroutines are finished wg.Wait() - // Close all files - for _, reader := range f.lastPollReaders { - reader.Close() - } - - f.lastPollReaders = readers - f.saveCurrent(readers) f.syncLastPollFiles(ctx) } @@ -251,8 +215,9 @@ func (f *InputOperator) makeReaders(filesPaths []string) []*Reader { fps = append(fps, fp) } - // Exclude any empty fingerprints - for i := 0; i < len(fps); i++ { + // Exclude any empty fingerprints or duplicate fingerprints to avoid doubling up on copy-truncate files +OUTER: + for i := 0; i < len(fps); { fp := fps[i] if len(fp.FirstBytes) == 0 { if err := files[i].Close(); err != nil { @@ -262,6 +227,25 @@ func (f *InputOperator) makeReaders(filesPaths []string) []*Reader { fps = append(fps[:i], fps[i+1:]...) files = append(files[:i], files[i+1:]...) } + + for j := 0; j < len(fps); j++ { + if i == j { + // Skip checking itself + continue + } + + fp2 := fps[j] + if fp.StartsWith(fp2) || fp2.StartsWith(fp) { + // Exclude + if err := files[i].Close(); err != nil { + f.Errorf("problem closing file", "file", files[i].Name()) + } + fps = append(fps[:i], fps[i+1:]...) + files = append(files[:i], files[i+1:]...) + continue OUTER + } + } + i++ } readers := make([]*Reader, 0, len(fps)) diff --git a/operator/builtin/input/file/file_test.go b/operator/builtin/input/file/file_test.go index 1724c475..663876b8 100644 --- a/operator/builtin/input/file/file_test.go +++ b/operator/builtin/input/file/file_test.go @@ -111,7 +111,6 @@ func TestReadExistingAndNewLogs(t *testing.T) { t.Parallel() operator, logReceived, tempDir := newTestFileOperator(t, nil, nil) operator.persister = testutil.NewMockPersister("test") - defer operator.Stop() // Start with a file with an entry in it, and expect that entry // to come through when we poll for the first time @@ -135,7 +134,6 @@ func TestStartAtEnd(t *testing.T) { cfg.StartAt = "end" }, nil) operator.persister = testutil.NewMockPersister("test") - defer operator.Stop() temp := openTemp(t, tempDir) writeString(t, temp, "testlog1\n") @@ -158,9 +156,9 @@ func TestStartAtEndNewFile(t *testing.T) { operator, logReceived, tempDir := newTestFileOperator(t, nil, nil) operator.persister = testutil.NewMockPersister("test") operator.startAtBeginning = false - defer operator.Stop() operator.poll(context.Background()) + temp := openTemp(t, tempDir) writeString(t, temp, "testlog1\ntestlog2\n") @@ -207,7 +205,6 @@ func TestSplitWrite(t *testing.T) { t.Parallel() operator, logReceived, tempDir := newTestFileOperator(t, nil, nil) operator.persister = testutil.NewMockPersister("test") - defer operator.Stop() temp := openTemp(t, tempDir) writeString(t, temp, "testlog1") @@ -432,11 +429,10 @@ func TestFileBatching(t *testing.T) { files := 100 linesPerFile := 10 - maxConcurrentFiles := 20 - maxBatchFiles := maxConcurrentFiles / 2 + maxConcurrentFiles := 10 - expectedBatches := files / maxBatchFiles // assumes no remainder - expectedLinesPerBatch := maxBatchFiles * linesPerFile + expectedBatches := files / maxConcurrentFiles // assumes no remainder + expectedLinesPerBatch := maxConcurrentFiles * linesPerFile expectedMessages := make([]string, 0, files*linesPerFile) actualMessages := make([]string, 0, files*linesPerFile) @@ -446,11 +442,10 @@ func TestFileBatching(t *testing.T) { cfg.MaxConcurrentFiles = maxConcurrentFiles }, func(out *testutil.FakeOutput) { - out.Received = make(chan *entry.Entry, expectedLinesPerBatch*2) + out.Received = make(chan *entry.Entry, expectedLinesPerBatch) }, ) operator.persister = testutil.NewMockPersister("test") - defer operator.Stop() temps := make([]*os.File, 0, files) for i := 0; i < files; i++ { @@ -470,6 +465,7 @@ func TestFileBatching(t *testing.T) { // poll once so we can validate that files were batched operator.poll(context.Background()) actualMessages = append(actualMessages, waitForN(t, logReceived, expectedLinesPerBatch)...) + expectNoMessagesUntil(t, logReceived, 10*time.Millisecond) } require.ElementsMatch(t, expectedMessages, actualMessages) @@ -487,6 +483,7 @@ func TestFileBatching(t *testing.T) { // poll once so we can validate that files were batched operator.poll(context.Background()) actualMessages = append(actualMessages, waitForN(t, logReceived, expectedLinesPerBatch)...) + expectNoMessagesUntil(t, logReceived, 10*time.Millisecond) } require.ElementsMatch(t, expectedMessages, actualMessages) @@ -496,7 +493,6 @@ func TestFileReader_FingerprintUpdated(t *testing.T) { t.Parallel() operator, logReceived, tempDir := newTestFileOperator(t, nil, nil) - defer operator.Stop() temp := openTemp(t, tempDir) tempCopy := openFile(t, temp.Name()) @@ -504,7 +500,6 @@ func TestFileReader_FingerprintUpdated(t *testing.T) { require.NoError(t, err) reader, err := operator.NewReader(temp.Name(), tempCopy, fp) require.NoError(t, err) - defer reader.Close() writeString(t, temp, "testlog1\n") reader.ReadToEnd(context.Background()) diff --git a/operator/builtin/input/file/reader.go b/operator/builtin/input/file/reader.go index a7b370c9..ac79bd5c 100644 --- a/operator/builtin/input/file/reader.go +++ b/operator/builtin/input/file/reader.go @@ -84,6 +84,12 @@ func (f *Reader) InitializeOffset(startAtBeginning bool) error { // ReadToEnd will read until the end of the file func (f *Reader) ReadToEnd(ctx context.Context) { + defer func() { + if err := f.file.Close(); err != nil { + f.Errorw("Failed to close", zap.Error(err)) + } + }() + if _, err := f.file.Seek(f.Offset, 0); err != nil { f.Errorw("Failed to seek", zap.Error(err)) return @@ -116,12 +122,8 @@ func (f *Reader) ReadToEnd(ctx context.Context) { } // Close will close the file -func (f *Reader) Close() { - if f.file != nil { - if err := f.file.Close(); err != nil { - f.Debugf("Problem closing reader", "Error", err.Error()) - } - } +func (f *Reader) Close() error { + return f.file.Close() } // Emit creates an entry with the decoded message and sends it to the next diff --git a/operator/builtin/input/file/rotation_test.go b/operator/builtin/input/file/rotation_test.go index 2b4a76f6..72a43b28 100644 --- a/operator/builtin/input/file/rotation_test.go +++ b/operator/builtin/input/file/rotation_test.go @@ -355,42 +355,6 @@ func TestMoveFile(t *testing.T) { expectNoMessages(t, logReceived) } -func TestTrackMovedAwayFiles(t *testing.T) { - if runtime.GOOS == "windows" { - t.Skip("Moving files while open is unsupported on Windows") - } - t.Parallel() - operator, logReceived, tempDir := newTestFileOperator(t, nil, nil) - operator.persister = testutil.NewMockPersister("test") - - temp1 := openTemp(t, tempDir) - writeString(t, temp1, "testlog1\n") - temp1.Close() - - operator.poll(context.Background()) - defer operator.Stop() - - waitForMessage(t, logReceived, "testlog1") - - // Wait until all goroutines are finished before renaming - operator.wg.Wait() - - newDir := fmt.Sprintf("%s%s", tempDir[:len(tempDir)-1], "_new/") - err := os.Mkdir(newDir, 0777) - require.NoError(t, err) - newFileName := fmt.Sprintf("%s%s", newDir, "newfile.log") - - err = os.Rename(temp1.Name(), newFileName) - require.NoError(t, err) - - movedFile, err := os.OpenFile(newFileName, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0644) - require.NoError(t, err) - writeString(t, movedFile, "testlog2\n") - operator.poll(context.Background()) - - waitForMessage(t, logReceived, "testlog2") -} - // TruncateThenWrite tests that, after a file has been truncated, // any new writes are picked up func TestTruncateThenWrite(t *testing.T) { diff --git a/operator/builtin/input/file/util_test.go b/operator/builtin/input/file/util_test.go index e1f8d167..eba5808f 100644 --- a/operator/builtin/input/file/util_test.go +++ b/operator/builtin/input/file/util_test.go @@ -122,7 +122,7 @@ func waitForOne(t *testing.T, c chan *entry.Entry) *entry.Entry { select { case e := <-c: return e - case <-time.After(3 * time.Second): + case <-time.After(time.Second): require.FailNow(t, "Timed out waiting for message") return nil } @@ -134,7 +134,7 @@ func waitForN(t *testing.T, c chan *entry.Entry, n int) []string { select { case e := <-c: messages = append(messages, e.Body.(string)) - case <-time.After(3 * time.Second): + case <-time.After(time.Second): require.FailNow(t, "Timed out waiting for message") return nil } @@ -146,7 +146,7 @@ func waitForMessage(t *testing.T, c chan *entry.Entry, expected string) { select { case e := <-c: require.Equal(t, expected, e.Body.(string)) - case <-time.After(3 * time.Second): + case <-time.After(time.Second): require.FailNow(t, "Timed out waiting for message", expected) } } @@ -158,7 +158,7 @@ LOOP: select { case e := <-c: receivedMessages = append(receivedMessages, e.Body.(string)) - case <-time.After(3 * time.Second): + case <-time.After(time.Second): break LOOP } }