Visitors¶

Namespace: dftracer::utils::utilities::composites::dft::visitors

source

class BloomVisitor : public dftracer::utils::utilities::composites::dft::DftEventVisitor ¶

Public Types

enum FixedBloom¶

Fixed bloom filter slots. Indices match DEFAULT_BLOOM_DIMENSIONS order: name, cat, pid, tid, hhash, fhash, shash.

Values:

enumerator BF_NAME¶

enumerator BF_CAT¶

enumerator BF_PID¶

enumerator BF_TID¶

enumerator BF_HHASH¶

enumerator BF_FHASH¶

enumerator BF_SHASH¶

enumerator BF_COUNT¶

enum FixedDim¶

Fixed dimension_stats slots. Superset of bloom dims plus pid_tid, ts, dur (which are observed for range stats but not hashed).

Values:

enumerator FD_NAME¶

enumerator FD_CAT¶

enumerator FD_PID¶

enumerator FD_TID¶

enumerator FD_PID_TID¶

enumerator FD_HHASH¶

enumerator FD_FHASH¶

enumerator FD_SHASH¶

enumerator FD_TS¶

enumerator FD_DUR¶

enumerator FD_COUNT¶

using HashResolutions = indexing::HashResolutions¶

using ChunkStatistics = indexing::ChunkStatistics ¶

using ChunkDimensionStats = indexing::ChunkDimensionStats ¶

using ChunkIndexerConfig = indexing::ChunkIndexerConfig ¶

Public Functions

BloomVisitor(ChunkIndexerConfig config, std::vector<std::string> dimensions)¶

BloomVisitor(const BloomVisitor&) = delete¶

BloomVisitor &operator=(const BloomVisitor&) = delete¶

BloomVisitor(BloomVisitor&&) noexcept = default¶

BloomVisitor &operator=(BloomVisitor&&) noexcept = default¶

virtual void begin(std::size_t num_checkpoints) override¶

virtual void on_checkpoint(std::size_t checkpoint_idx) override¶

virtual void on_event(const EventRecord &record) override¶

virtual std::unique_ptr<DftEventVisitor> create_parallel_slice() const override¶

virtual void merge_parallel_slice(DftEventVisitor &slice) override¶

void finalize(indexer::IndexDatabaseWriterContext &writer, int file_id)¶

void finalize_sink_only(indexer::IndexBatchSink &sink, int file_id)¶: Emit bloom / stats / dimension records plus name dictionary/postings to a sink backend. Skips ROOT_* summaries (rebuilt separately by IndexDatabase::rebuild_root_summaries()). Works for both the RocksDB-backed writer and the SST writer.

void flush_per_checkpoint_to_sink(indexer::IndexBatchSink &sink, int file_id)¶: Emit per-checkpoint chunk records (bloom, stats, dim_stats, name_chunk_postings) using the current chunks_ buffer, merge their state into the persistent file-level accumulator, then clear chunks_ and advance the base index. Used for mid-chunk slice rotation when chunks_ would otherwise grow unbounded.

void finalize_file_to_sink(indexer::IndexBatchSink &sink, int file_id)¶: Emit file-level records (file_bloom, scalar_stats, counts, dimensions, name_dictionary, name_file_postings) from the persistent accumulator. Call once at end-of-file.

inline std::size_t num_chunks() const¶

inline std::uint64_t total_events() const¶: Total event count across already-flushed chunks plus the currently buffered ones. Reflects all events ingested via on_event() so far.

struct ChunkState¶

Public Functions

ChunkState()¶

Public Members

std::array<indexing::BloomFilter, BF_COUNT> fixed_blooms¶

std::array<ChunkDimensionStats, FD_COUNT> fixed_dim_stats¶

std::vector<indexing::BloomFilter> extra_blooms¶

std::vector<ChunkDimensionStats> extra_dim_stats¶

ChunkStatistics statistics¶

HashResolutions hash_resolutions¶

std::size_t events_processed = 0¶

source

class HashTableVisitor : public dftracer::utils::utilities::composites::dft::DftEventVisitor ¶

Captures FH/HH/SH/PR metadata events during indexing and stores them in HASH_TABLES column family with bidirectional lookups:

Forward (hash -> name): for resolving hashes in output
Reverse (name -> hash): for query DSL like file_name == "/path/..."

Public Types

enum class HashType : std::uint8_t¶

Hash table types matching dfanalyzer naming conventions.

Values:

enumerator FILE¶

enumerator HOST¶

enumerator STRING¶

enumerator PROC¶

Public Functions

HashTableVisitor() = default¶

HashTableVisitor(const HashTableVisitor&) = delete¶

HashTableVisitor &operator=(const HashTableVisitor&) = delete¶

HashTableVisitor(HashTableVisitor&&) noexcept = default¶

HashTableVisitor &operator=(HashTableVisitor&&) noexcept = default¶

virtual void begin(std::size_t num_checkpoints) override¶

virtual void on_checkpoint(std::size_t checkpoint_idx) override¶

virtual void on_event(const EventRecord &record) override¶

virtual std::unique_ptr<DftEventVisitor> create_parallel_slice() const override¶

virtual void merge_parallel_slice(DftEventVisitor &slice) override¶

void finalize(indexer::IndexBatchSink &writer, int file_id)¶

std::size_t num_entries() const¶

source

class ManifestVisitor : public dftracer::utils::utilities::composites::dft::DftEventVisitor ¶

Public Functions

ManifestVisitor() = default¶

virtual void begin(std::size_t num_checkpoints) override¶

virtual void on_checkpoint(std::size_t checkpoint_idx) override¶

virtual void on_event(const EventRecord &record) override¶

virtual std::unique_ptr<DftEventVisitor> create_parallel_slice() const override¶

virtual void merge_parallel_slice(DftEventVisitor &slice) override¶

inline virtual void set_line_offset(std::size_t offset) override¶: In parallel-flush mode, slices receive events with slice-local line numbers (0..N-1). The dispatcher calls this on the slice before merge_parallel_slice with the cumulative successful-event count of prior slices, so the slice can renumber its stored line indices.

inline virtual std::size_t parallel_event_count() const override¶: Successful events processed by this slice. Used by the dispatcher to propagate line offsets across slices in byte order.

void finalize(indexer::IndexBatchSink &writer, int file_id)¶

void flush_per_checkpoint_to_sink(indexer::IndexBatchSink &sink, int file_id)¶: Emit per-checkpoint event/metadata line records and clear the vectors. Used for mid-chunk slice rotation.

void finalize_file_to_sink(indexer::IndexBatchSink &sink, int file_id)¶: Emit file-level records (observed pids). Call once at end-of-file.