Expand description
Compile-time and runtime constants for tuning, timeouts, and thresholds. Compile-time and runtime constants for foxing subsystems. Includes worker tuning, retry limits, buffer sizes, and feature flags.
Constants§
- AI_
BALANCED_ ARCHIVE_ PEEK_ BUDGET_ MS - Archive peek budget at Balanced effort.
- AI_
BALANCED_ MAX_ CONTENT_ CHARS - Maximum content chars at Balanced effort.
- AI_
BALANCED_ MAX_ PREVIEW_ CHARS - Max preview chars at Balanced effort.
- AI_
BALANCED_ MAX_ PROMPT_ CHARS - Fallback max prompt chars at Balanced effort.
- AI_
BINARY_ CAPSTONE_ PEEK_ BYTES - Bytes of entry-point code to disassemble via capstone.
- AI_
BINARY_ DEADLINE_ MS - Hard deadline for binary metadata extraction (milliseconds).
- AI_
BINARY_ MAX_ STRINGS - Maximum printable strings to extract from binary
.rodata/.rdata. - AI_
BINARY_ MAX_ SYMBOLS - Maximum symbols (imports + exports) before truncating.
- AI_
BINARY_ MIN_ STRING_ LEN - Minimum string length for binary string extraction.
- AI_
CHUNK_ OVERLAP_ CODE - Overlap for code chunks (chars).
- AI_
CHUNK_ OVERLAP_ DEFAULT - Default overlap for unknown content (chars).
- AI_
CHUNK_ OVERLAP_ DOCUMENT - Overlap for document chunks (chars).
- AI_
CHUNK_ OVERLAP_ MARKUP - Overlap for markup chunks (chars).
- AI_
CHUNK_ OVERLAP_ TEXT - Overlap for text chunks (chars).
- AI_
CHUNK_ PREVIEW_ CHARS - Default character limit for chunk content previews stored in
ChunkMeta, pgvectorfoxing_chunks.preview, and SSE payloads. - AI_
CHUNK_ SIZE_ CODE - Chunk size for code files (chars). Smaller chunks preserve function boundaries.
- AI_
CHUNK_ SIZE_ DEFAULT - Default chunk size for unknown/unrecognized content (chars).
- AI_
CHUNK_ SIZE_ DOCUMENT - Chunk size for document files like PDF/DOCX (chars). Large to preserve document structure.
- AI_
CHUNK_ SIZE_ MARKUP - Chunk size for markup files like HTML/XML (chars).
- AI_
CHUNK_ SIZE_ TEXT - Chunk size for text/prose files (chars). Larger to preserve paragraph context.
- AI_
COLLISION_ CONFIDENCE_ MAX - Maximum effective confidence after collision gate adjustment (ceiling to avoid 1.0).
- AI_
COLLISION_ CONFIDENCE_ MULTIPLIER - Confidence multiplier for extensions in COLLISION_MAP table. Effective threshold = AI_MAGIKA_CONFIDENCE_THRESHOLD * AI_COLLISION_CONFIDENCE_MULTIPLIER. At defaults: 0.50 * 1.5 = 0.75 effective threshold for collision-prone extensions. This matches Magika’s own raised threshold for markdown (0.75 in config.min.json).
- AI_
CONTAINER_ MAX_ FILES_ DISPLAY - Maximum file entries displayed in container sidecar context.
- AI_
CONTAINER_ MAX_ FILE_ ENTRIES - Maximum file entries to extract from RPM metadata.
- AI_
CONTAINER_ MAX_ OCI_ LAYERS - Maximum OCI manifest layers to extract.
- AI_
CONTAINER_ MAX_ REQUIRES - Maximum requires to extract from RPM metadata.
- AI_
CONTAINER_ MAX_ REQUIRES_ DISPLAY - Maximum requires entries displayed in container sidecar context.
- AI_
DEEP_ ARCHIVE_ PEEK_ BUDGET_ MS - Archive peek budget at Deep effort (0 = no budget cap).
- AI_
DEEP_ MAX_ CONTENT_ CHARS - Maximum content chars at Deep effort.
- AI_
DEEP_ MAX_ PREVIEW_ CHARS - Max preview chars at Deep effort.
- AI_
DEEP_ MAX_ PROMPT_ CHARS - Fallback max prompt chars at Deep effort.
- AI_
DEEP_ MAX_ PROMPT_ CHARS_ CAP - Hard cap on prompt chars for deep effort with probed context (chars).
- AI_
DEFAULT_ EMBEDDING_ DIMS - Default embedding vector dimensions used when the model path is unknown or empty (HTTP-only mode).
- AI_
DIR_ CONTEXT_ MIN_ DOMINANT_ FRACTION - Minimum fraction (0.0–1.0) of files sharing the dominant extension before
format_directory_context()emits a directory context string. Below this threshold the directory is too heterogeneous — context would be misleading noise rather than useful signal for the LLM. Example: 10 files, 3 .gg + 2 .zip + 2 .png + 3 others → dominant = 30% → suppressed. 100 files, 85 .smc + 10 .txt + 5 .xml → dominant = 85% → emitted. - AI_
DOCLING_ POLL_ INTERVAL_ MS - Poll interval for docling-serve async status checks (milliseconds).
- AI_
DOCLING_ TIMEOUT_ SECS - Maximum wait time for async docling-serve conversion (seconds).
- AI_
HNSW_ CAPACITY_ GROWTH_ INCREMENT - HNSW capacity growth increment when the index is full.
- AI_
HNSW_ EF_ CONSTRUCTION_ COSINE - HNSW
ef_constructionfor embedding cosine-similarity index. - AI_
HNSW_ EF_ CONSTRUCTION_ SIMHASH - HNSW
ef_constructionfor 64-bit SimHash binary index (bit_hamming_ops). Set to 256 per SOTA recommendation for binary vectors with high near-duplicate ratios (NapierOne dataset analysis: 64 produces poor recall on dense similarity clusters). - AI_
HNSW_ M - HNSW
mparameter – max bi-directional links per node. - AI_
LLM_ BUDGET_ RESET_ SECS - Budget reset interval – tokens_used counter resets after this many seconds.
- AI_
LLM_ DAILY_ TOKEN_ BUDGET - Default daily token budget (input + output). 0 = unlimited. Offline-first default: local LLM servers (llama.cpp, vLLM, Ollama) have no token cost. Set to a non-zero value only for paid cloud APIs (e.g. 100_000 for OpenAI free tier).
- AI_
LLM_ DEFAULT_ MAX_ TOKENS - Default max output tokens per LLM enrichment request.
Applies when
ai.enrichment.max_tokensis unset. - AI_
LLM_ DEFAULT_ TEMPERATURE - Default sampling temperature for LLM enrichment requests. Low value (0.2) favours consistency over creativity.
- AI_
LLM_ MAX_ CONCURRENT - Maximum concurrent LLM enrichment requests.
- AI_
LLM_ MAX_ RETRIES - Maximum retries on transient LLM failures before giving up.
- AI_
LLM_ REQUEST_ TIMEOUT_ SECS - HTTP timeout per LLM enrichment request (seconds).
- AI_
LLM_ RETRYABLE_ STATUS_ CODES - HTTP status codes that indicate a retryable LLM API error.
- AI_
LLM_ RETRY_ INITIAL_ DELAY_ MS - Initial backoff delay for LLM retry loop (milliseconds).
- AI_
LLM_ RETRY_ MAX_ DELAY_ MS - Maximum backoff delay for LLM retry loop (milliseconds).
- AI_
LLM_ RETRY_ MIN_ DELAY_ MS - Minimum retry delay for LLM API calls in milliseconds (floor for exponential backoff).
- AI_
LLM_ RPM_ WINDOW_ SECS - RPM sliding window duration for LLM rate limiting (seconds).
- AI_
MAGIKA_ CONFIDENCE_ THRESHOLD - Default Magika ML confidence threshold for the confidence gate in
MagikaRouter::classify_file()andmedia_meta::extract_metadata(). At or above this score, Magika’s classification is trusted directly. Below this score, the gate checks whether the file extension is inMAGIKA_KNOWN_EXTENSIONS: if yes, trust Magika; if no, fall back to extension-basedContentRouter. Configurable via[ai.enrichment] magika_confidence_thresholdin foxingd config. CLIfxcp enrichuses this constant as the default. - AI_
MAGIKA_ HIGH_ CONFIDENCE - Magika ML confidence threshold for full trust (ignore extension). At or above this, Magika’s label/group/is_text are used directly.
- AI_
MAGIKA_ SUPPLEMENT_ THRESHOLD - Magika ML confidence threshold for extension-supplemented routing. Between this and AI_MAGIKA_HIGH_CONFIDENCE, Magika is used but extension info supplements when there’s a conflict. Below this threshold, fall back to extension-only classification.
- AI_
MAX_ ASSOCIATION_ CONTEXT_ CHARS - Maximum chars of association context (Dublin Core descriptions from associated files) injected into the enrichment prompt. Truncated at UTF-8 boundary.
- AI_
MAX_ COLLECTION_ CONTEXT_ CHARS - Maximum chars for the collection context section. Default: 500 chars.
- AI_
MAX_ CONTENT_ CHARS - Maximum chars of raw file content in the enrichment prompt. This is the unstructured content budget AFTER metadata context. Truncated at UTF-8 boundary. Default: 4000 chars.
- AI_
MAX_ EMBED_ BYTES - Maximum bytes of text content to embed via chunking.
Content beyond this point is truncated before
split_recursive()to bound ORT embedding cost while still allowing large files to be indexed. 1 MiB of UTF-8 text produces ~4K chunks at 256 chars each. - AI_
MAX_ KEYWORDS_ PER_ FILE - Maximum number of keywords extracted per file in LLM enrichment output.
- AI_
MAX_ MENTIONS_ PER_ FILE - Maximum number of entity mentions extracted per file in LLM enrichment output.
- AI_
MAX_ PREVIEW_ CHARS - Maximum chars for the indexed content preview section. Prevents large multi-chunk files from blowing out the prompt. Default: 1000 chars.
- AI_
MAX_ PROMPT_ CHARS - Maximum total prompt length (chars) sent to the LLM for enrichment. Metadata context (Magika, extracted metadata, similar files) is always included first. Raw file content fills the remaining budget. Default: 12000 chars (~=3000 tokens) – leaves headroom for LLM response within a 16K context window.
- AI_
MAX_ REFERENCE_ CONTEXT_ CHARS - Maximum chars of reference file context injected per enrichment prompt.
- AI_
MAX_ SIDECAR_ CONTEXT_ CHARS - Maximum chars of sidecar context injected into the enrichment prompt.
- AI_
MAX_ TEXT_ FILE_ SIZE - 100 MB – applies only to text/code files. Header-based groups (media, video, archive, binary, document) have no size limit.
- AI_
MEDIA_ ASSOC_ EXACT_ CONFIDENCE - Confidence score for exact stem-match media-to-parent file association.
- AI_
MEDIA_ ASSOC_ FUZZY_ CONFIDENCE_ MULTIPLIER - Confidence multiplier for fuzzy media file association scores.
- AI_
MEDIA_ ASSOC_ FUZZY_ THRESHOLD - Jaro-Winkler similarity threshold for fuzzy media file association.
- AI_
ORT_ DEFAULT_ BATCH_ SIZE - Default batch size for ORT embedding inference.
- AI_
ORT_ MAX_ INTRA_ THREADS - Maximum ORT intra-op parallelism threads. Caps thread count to avoid over-subscribing CPU cores on high-core-count machines.
- AI_
ORT_ MAX_ SEQUENCE_ LENGTH - Maximum input sequence length (tokens) for ORT/ONNX tokenizer.
- AI_
PGVECTOR_ CHUNK_ SUB_ BATCH - SQL parameter limit chunk size for pgvector batch inserts.
- AI_
PGVECTOR_ FILE_ BATCH_ SIZE - File batch size per transaction during pgvector export.
- AI_
PGVECTOR_ FLUSH_ BATCH_ SIZE - Default flush batch size for pgvector sync.
- AI_
PGVECTOR_ MAX_ CHUNK_ PREVIEWS - Maximum chunk content previews returned per file from pgvector.
- AI_
PGVECTOR_ MAX_ QUEUE_ DEPTH - Default maximum pgvector queue depth before back-pressure.
- AI_
PGVECTOR_ PULL_ MODE_ AUTHORITATIVE - CID pull-through mode: accept any enrichment for matched CIDs, never call LLM.
- AI_
PGVECTOR_ PULL_ MODE_ OPPORTUNISTIC - CID pull-through mode: accept enrichment only when stored effort >= requested effort.
- AI_
PGVECTOR_ RECONNECT_ BASE_ SECS - Base interval for pgvector reconnection attempts when initial connection fails (seconds).
- AI_
PGVECTOR_ RECONNECT_ MAX_ SECS - Maximum interval for pgvector reconnection with exponential backoff cap (seconds).
- AI_
PICO8_ DETECTION_ CONFIDENCE - Confidence score assigned to files detected as Pico-8 cartridges via dimension probe.
- AI_
PICO8_ PNG_ HEIGHT - Pico-8 cartridge PNG height.
- AI_
PICO8_ PNG_ WIDTH - Pico-8 cartridge PNG width (steganographic game data embedded in pixels).
- AI_
QUICK_ ARCHIVE_ PEEK_ BUDGET_ MS - Archive peek budget (milliseconds) per effort level.
- AI_
QUICK_ MAX_ CONTENT_ CHARS - Maximum content chars sent to LLM at Quick effort. Quick skips LLM entirely, but if forced: minimal content.
- AI_
QUICK_ MAX_ PREVIEW_ CHARS - Max preview chars at each effort level.
- AI_
QUICK_ MAX_ PROMPT_ CHARS - Fallback max prompt chars at Quick effort.
- AI_
REFERENCE_ LLM_ THRESHOLD - LLM-layer confidence threshold for reference file injection.
- AI_
REFERENCE_ ORT_ THRESHOLD - ORT-layer confidence threshold for reference file injection. At >=0.90, reference file context can be injected without LLM gating.
- AI_
RRF_ K - Reciprocal Rank Fusion (RRF) parameter
kfor hybrid search scoring. Higher k smooths rank differences; k=60 is the standard default (Cormack et al.). - AI_
SAFE_ TRUNCATE_ ABSTRACT_ LIMIT - Truncation limit for dc_abstract when LLM returns non-JSON (chars).
- AI_
SAFE_ TRUNCATE_ CONTEXT_ LIMIT - Truncation limit for sidecar context and extracted metadata in synthesised enrichment prompts (chars).
- AI_
SAFE_ TRUNCATE_ DESCRIPTION_ LIMIT - Truncation limit for description when LLM returns non-JSON (chars).
- AI_
SANDWICH_ HEAD_ RATIO - Head section ratio for sandwich sampling (35% of budget).
- AI_
SANDWICH_ MARKER_ HEAD - Marker prefix for the head section in sandwich output.
- AI_
SANDWICH_ MARKER_ MID - Marker prefix for the middle section in sandwich output.
- AI_
SANDWICH_ MARKER_ OVERHEAD - Maximum byte overhead from section markers and newlines.
- AI_
SANDWICH_ MARKER_ SAMPLE - Marker prefix for random sample sections in sandwich output.
- AI_
SANDWICH_ MARKER_ TAIL - Marker prefix for the tail section in sandwich output.
- AI_
SANDWICH_ MID_ RATIO - Middle section ratio for sandwich sampling (25% of budget).
- AI_
SANDWICH_ RANDOM_ RATIO - Random sample ratio for sandwich sampling (20% of budget).
- AI_
SANDWICH_ RANDOM_ WINDOWS - Number of random sample windows extracted from the interior.
- AI_
SANDWICH_ TAIL_ RATIO - Tail section ratio for sandwich sampling (20% of budget).
- AI_
SIDECAR_ ANCESTOR_ MAX_ DEPTH - Maximum ancestor directory depth for gamelist.xml walk-up discovery.
- AI_
SIDECAR_ CONFIDENCE_ THRESHOLD - Minimum confidence to pass sidecar data to the LLM layer for evaluation. LLM can reason about borderline cases; below this threshold, discard.
- AI_
SIDECAR_ EXACT_ CONFIDENCE - Confidence score for exact filename-stem sidecar matches.
- AI_
SIDECAR_ EXACT_ CONTEXT_ AFTER - Context lines after an exact filename-stem match (i + N).
- AI_
SIDECAR_ EXACT_ CONTEXT_ BEFORE - Context lines before an exact filename-stem match.
- AI_
SIDECAR_ EXACT_ SNIPPET_ CHARS - Exact match snippet truncation (chars).
- AI_
SIDECAR_ FUZZY_ CONFIDENCE_ MULTIPLIER - Confidence multiplier applied to Jaro-Winkler score for fuzzy sidecar matches.
- AI_
SIDECAR_ FUZZY_ CONTEXT_ AFTER - Context lines after a fuzzy filename-stem match (idx + N).
- AI_
SIDECAR_ FUZZY_ CONTEXT_ BEFORE - Context lines before a fuzzy filename-stem match.
- AI_
SIDECAR_ FUZZY_ SNIPPET_ CHARS - Fuzzy match snippet truncation (chars).
- AI_
SIDECAR_ FUZZY_ THRESHOLD - Jaro-Winkler similarity threshold for fuzzy sidecar filename matching.
- AI_
SIDECAR_ HIGH_ CONFIDENCE - Minimum Magika confidence to accept ORT-layer sidecar data without LLM review. At >=0.85, ORT alone can make the accept/reject decision.
- AI_
SIDECAR_ KV_ CONFIDENCE - Confidence score for structured key-value sidecar matches.
- AI_
SIDECAR_ KV_ MAX_ LINES - Maximum KV-style lines to extract from a structured sidecar.
- AI_
SIDECAR_ KV_ MAX_ LINE_ LEN - Maximum line length for key-value pairs in raw sidecar parsing.
- AI_
SIDECAR_ KV_ MIN_ LINES - Minimum KV-style lines to classify a sidecar as structured.
- AI_
SIDECAR_ LOW_ ADOPTION_ THRESHOLD - Adoption threshold: if fewer than 30% of files in a collection have a given sidecar format, treat it as low-adoption and apply stricter confidence gating.
- AI_
SIDECAR_ MAX_ SOURCES_ PER_ FILE - Maximum number of sidecar sources attached per file enrichment.
- AI_
SIDECAR_ MIN_ DIR_ FILES - Minimum files in a directory before sidecar detection activates.
- AI_
SIDECAR_ RAW_ CONFIDENCE - Confidence score for raw/unstructured sidecar fallback matches.
- AI_
SIDECAR_ RAW_ SNIPPET_ CHARS - Raw fallback snippet truncation (chars).
- AI_
SIDECAR_ STALENESS_ PENALTY_ RATIO - Staleness penalty ratio: sidecar data older than Nx the file’s own mtime delta gets a confidence penalty applied during sidecar scoring.
- AI_
SIDECAR_ XML_ CONFIDENCE - Confidence score for XML-structured sidecar matches (highest structural signal).
- AI_
SIDECAR_ XML_ DEPTH_ PENALTY - Depth penalty multiplier applied to XML sidecar confidence scores.
- AI_
SIDECAR_ XML_ SNIPPET_ CHARS - XML match snippet truncation (chars).
- AI_
SWEEP_ MAX_ BACKOFF_ MULTIPLIER - Maximum multiplier for exponential backoff on empty AI sweeps. When a periodic sweep finds zero files needing work, the interval doubles (up to base_interval × this cap). Resets to base on any non-empty sweep.
- AI_
TAXONOMY_ NEAREST_ THRESHOLD - Minimum cosine similarity for taxonomy_nearest() pgvector HNSW fallback to populate PipelineContext.
- AI_
TOKENS_ TO_ CHARS_ RATIO - Approximate characters-per-token ratio for LLM token budget estimation. Used to convert available token budget into a character limit for content truncation.
- API_
LISTEN_ BACKLOG - TCP listen backlog for the Prometheus/API server socket.
- AVG_
EVENT_ OVERHEAD_ BYTES - Average total memory cost per queued event, including payload.
- BARRIER_
BROADCAST_ CHANNEL_ CAPACITY - Broadcast channel capacity for barrier coordination.
- BARRIER_
CHECK_ INTERVAL - Polling interval (ms) for WAL barrier completion checks.
- BPF_
DEDUP_ CLEANUP_ INTERVAL - Periodic cleanup interval for the dedup cache (every N events).
- BPF_
DEDUP_ RETAIN_ FACTOR - How many dedup windows to retain before evicting (cleanup keeps Nx window).
- BPF_
DEDUP_ WINDOW_ MS - Dedup window for kprobe/fentry/LSM duplicate events (milliseconds).
- BPF_
EVENT_ OVERHEAD_ BYTES - Per-event fixed overhead estimate (bytes) used by the ReorderBuffer for memory accounting alongside the event payload size.
- BPF_
EVENT_ SIZE_ ESTIMATE_ BYTES - Base event payload size estimate (bytes) used by the ReorderBuffer for memory accounting. The actual size adds the event name length.
- BPF_
EVENT_ VERSION - Wire format version for BPF ring buffer event structs.
- BPF_
MAX_ TRACKED_ DEVICES - Maximum number of block devices tracked simultaneously in the BPF sequence tracker and identity map.
- BPF_
MAX_ TRANSIENT_ ENTRIES - Maximum entries in the TransientFilter map before forced GC + eviction. Prevents unbounded HashMap growth under sustained create storms.
- BPF_
READY_ TIMEOUT_ SECS - Timeout waiting for BPF probes to attach at daemon startup (seconds).
- BPF_
RING_ POLL_ INTERVAL_ MS - BPF ring buffer poll interval (milliseconds). Controls how often the userspace event loop polls the kernel ring buffer for new events.
- BPF_
SEQUENCE_ GAP_ THRESHOLD - Sequence gap threshold for BPF ring buffer. Gaps larger than this trigger panic-mode hydration and a synthetic SequenceGap event.
- BPF_
STATS_ REPORT_ INTERVAL_ SECS - Interval between BPF per-device statistics reports (seconds).
- BPF_
TRANSIENT_ GC_ CUTOFF_ SECS - TransientFilter GC cutoff (seconds). Entries older than this are evicted during periodic garbage collection to prevent unbounded growth.
- BROWSER_
EVENT_ POLL_ INTERVAL_ MS - TUI browser event poll interval (milliseconds).
- BTRFS_
SUPER_ MAGIC - btrfs filesystem magic number (from statfs).
- BUFFER_
POOL_ MAX_ REGISTERED_ BUFFERS - Maximum number of registered buffers in a BufferPool.
- BUFFER_
POOL_ MIN_ CHUNK_ SIZE - Smallest allowed chunk size in the buffer pool (bytes).
- CACHE_
RESIDENCY_ SENDFILE_ THRESHOLD - Minimum page-cache residency ratio to trigger the cachestat sendfile fast-path. At >=80% resident, sendfile outperforms io_uring for files in the decision zone.
- CACHE_
TTL_ EVENTS - Number of events processed before capability caches are invalidated.
- CAPACITY_
THRESHOLD_ MB - Free-space threshold on the target: below this, the daemon warns about low capacity.
- CHUNK_
READER_ MAX_ INPUT - Maximum input size for the chunk_reader function (16MB).
- CHUNK_
READER_ READ_ BUFFER_ SIZE - Read buffer size for chunk_reader.
- CIFS_
MAGIC_ NUMBER - CIFS filesystem magic number (from statfs).
- COALESCER_
BATCH_ INITIAL_ CAPACITY - Initial Vec capacity for EventBatch in the Coalescer.
- COALESCER_
FRONTIER_ WIDTH_ LIMIT - Bounded frontier width limit for the Coalescer. When the batch exceeds this threshold, aggressive transient lifecycle pruning is triggered.
- COALESCER_
MAX_ CAPACITY - Maximum capacity of the per-worker write coalescer. Also used as the denominator for buffer utilization metrics.
- CONTENT_
VERIFY_ SAMPLE_ SIZE - Sample size for head/tail content verification (64KB).
- CONTENT_
VERIFY_ TAIL_ THRESHOLD - Minimum file size before tail sampling is used during content verification (128KB).
- COPY_
FILE_ RANGE_ CHUNK_ SIZE - Maximum chunk size per
copy_file_range(2)call (1 GiB). Kernel may copy less; caller loops until complete. - COPY_
TIMEOUT_ BASE_ SECS - Base timeout for copy operations (seconds). The actual timeout is computed
as
base + (file_size / 10MB) * per_10mb, capped atCOPY_TIMEOUT_CAP_SECS. - COPY_
TIMEOUT_ CAP_ SECS - Maximum copy timeout regardless of file size (seconds).
- COPY_
TIMEOUT_ LARGE_ SECS - Timeout for copying large files. Allows for slow network or HDD targets.
- COPY_
TIMEOUT_ PER_ 10MB_ SECS - Additional timeout per 10 MB of file data (seconds).
- COPY_
TIMEOUT_ SMALL_ SECS - Timeout for copying small files (< 64KB). Covers metadata-heavy NFS round-trips.
- DAEMON_
PROBE_ TIMEOUT_ SECS - HTTP probe timeout for
is_daemon_runningliveness check (seconds). - DEFAULT_
HIBERNATION_ SECS - Seconds of inactivity before the daemon enters hibernation (reduced polling).
- DEFAULT_
IDENTITY_ MAP_ OVERHEAD_ BYTES - Estimated per-entry memory overhead for the inode identity map (bytes).
- DEFAULT_
IO_ BUFFER_ SIZE_ MIB - Default I/O buffer size for copy operations (MiB).
- DEFAULT_
METRICS_ PORT - Default Prometheus metrics port for the foxingd API server.
- DEFAULT_
QUEUE_ MAX - Default maximum event queue depth before back-pressure engages.
- DELTA_
GOOD_ MATCH_ LEN - Match length considered “good enough” to stop early search in delta encoding.
- DELTA_
MAX_ INPUT_ SIZE - Maximum input size for delta encoding (64KB).
- DELTA_
MIN_ MATCH_ LEN - Minimum match length for a delta Copy operation.
- DIRTY_
JOURNAL_ MAX_ BYTES - Maximum dirty-range journal file size before the journal is discarded and a full Merkle resync is triggered (bytes). 10 MiB.
- DIRTY_
RANGE_ MAX_ PER_ FILE - Maximum dirty ranges tracked per file before overflow to full resync.
- DISPATCHER_
BULK_ TIN_ CAPACITY - Channel capacity for the bulk tin in the TinnedDispatcher. Bulk events (write, truncate) are freely droppable; hydration catches misses.
- DISPATCHER_
METADATA_ TIN_ CAPACITY - Channel capacity for the metadata tin in the TinnedDispatcher. Metadata events (chmod, chown, xattr) are droppable under pressure.
- DRAIN_
ACK_ TIMEOUT_ SECS - Timeout for workers to acknowledge a drain barrier during recovery (seconds).
- EMERGENCY_
PRUNE_ THRESHOLD_ BYTES - Emergency prune threshold (bytes). When target storage drops below this, the worker initiates emergency version pruning before retrying the copy.
- ERROR_
LIMITER_ SECS - Minimum interval between repeated error log messages for the same path.
- ESTIMATED_
RING_ DEPTH - Estimated number of in-flight entries in the BPF ring buffer.
- EVENTBUS_
NOTIFY_ CHANNEL_ CAPACITY - Broadcast channel capacity for EventBus change notifications.
- EVENT_
HANDLER_ TIMEOUT_ SECS - Timeout for simple event handler operations: rename, truncate, fallocate (seconds).
- EVENT_
QUEUE_ OVERHEAD_ BYTES - Per-event memory overhead in the dispatcher queue (bytes).
- F2FS_
SUPER_ MAGIC - F2FS filesystem magic number (from statfs).
- FICLONE_
IOCTL - FICLONE ioctl number for CoW reflink cloning.
- FSYNC_
LATENCY_ DEFAULT_ AVG_ US - Default average fsync latency estimate for
FsyncLatencyTracker(microseconds). Corresponds to 1 second – conservative initial estimate for cold starts. - FSYNC_
LATENCY_ DEFAULT_ DEVIATION_ US - Default fsync latency deviation for
FsyncLatencyTracker(microseconds). Used in adaptive timeout calculation: timeout = avg + 4 x deviation. - FSYNC_
LATENCY_ MAX_ US - Maximum fsync latency estimate after timeout doubling (microseconds).
Caps
record_timeout()to prevent runaway growth (30 seconds). - FSYNC_
TIMEOUT_ MAX_ US - Maximum adaptive fsync timeout (microseconds). Ceiling for
get_timeout()clamp (60 seconds). - FSYNC_
TIMEOUT_ MIN_ US - Minimum adaptive fsync timeout (microseconds). Floor for
get_timeout()clamp (5 seconds). - FS_
IOC_ SETFLAGS_ IOCTL - FS_IOC_SETFLAGS ioctl for setting filesystem flags (immutable, append-only).
- FULL_
SCAN_ DEBOUNCE_ SECS - Minimum quiet period between consecutive full-scan triggers.
- FULL_
SCAN_ EPOCH_ OFFSET_ SECS - Epoch offset for full-scan debounce:
last_full_scanis initialized tonow - offsetso the first health-check-triggered scan fires immediately. - FUSE_
CACHE_ TTL_ DEFAULT_ SECS - Default metadata cache TTL (seconds) for the FUSE passthrough overlay.
StorageClassification-aware overrides apply at runtime: NFS 30s, HDD 60s, NVMe 5s.
Corresponds to
fuse_cache_ttl_secsinTargetConfig. - FXAR_
RESTORE_ HEADROOM_ BYTES - Headroom bytes reserved for FXAR restore disk space checks (10MB).
- FXCP_
BUFFER_ POOL_ COUNT - Buffer pool count for fxcp CLI tree-copy io_uring pipeline.
- FXCP_
IO_ URING_ SQ_ DEPTH - io_uring submission queue depth for fxcp CLI tree-copy operations.
- GC_
INTERVAL_ SECS - Interval between garbage collection sweeps for expired map entries.
- GC_
STARTUP_ JITTER_ MAX_ MS - Maximum random jitter added to GC startup to avoid thundering herd (ms).
- GENERATION_
UNKNOWN - Sentinel value for unknown/unset inode generation.
- GIT_
COMMIT_ HASH - Git commit hash at build time, or “unknown” if not set.
- GOVERNOR_
CHECK_ INTERVAL_ MS - How often the governor samples PSI, memory, and CPU pressure (ms).
- GOVERNOR_
EMERGENCY_ PRESSURE_ THRESHOLD - Emergency pressure threshold for AI enrichment skip.
- GOVERNOR_
FAILURE_ RATE_ THRESHOLD - Copy failure rate (0.0..1.0) above which the governor declares the target unhealthy.
- GOVERNOR_
FAILURE_ WINDOW_ RESET_ OPS - Total operations before the governor auto-resets its failure window. Prevents transient error bursts from permanently marking a target unhealthy.
- GOVERNOR_
FAILURE_ WINDOW_ SECS - Sliding window for computing the governor’s failure rate.
- GOVERNOR_
HYPERVISOR_ PSI_ MULTIPLIER - PSI relaxation multiplier for hypervisor/container environments. Hypervisor-reported PSI values are inflated by host contention; this multiplier relaxes the threshold to avoid false throttling.
- GOVERNOR_
HYSTERESIS_ SECS - Cooldown period before the governor can toggle stress state again.
- GOVERNOR_
MEMORY_ HIGH_ WATERMARK_ PCT - Memory usage fraction above which the governor activates pressure throttling.
- GOVERNOR_
PACING_ INITIAL_ MS - Initial pacing delay when the governor first detects system stress (ms).
- GOVERNOR_
PACING_ MAX_ MS - Maximum pacing delay the governor can impose on workers (ms).
- GOVERNOR_
PSI_ CPU_ THRESHOLD_ NORMAL - PSI CPU pressure percentage that triggers governor throttling (normal mode).
- GOVERNOR_
PSI_ CPU_ THRESHOLD_ RELAXED - PSI CPU pressure threshold for hypervisor/container environments (relaxed mode).
- GOVERNOR_
PSI_ IO_ THRESHOLD_ NORMAL - PSI I/O pressure percentage that triggers governor throttling (normal mode).
- GOVERNOR_
PSI_ IO_ THRESHOLD_ RELAXED - PSI I/O pressure threshold for hypervisor/container environments (relaxed mode).
- GOVERNOR_
THROTTLE_ PRESSURE_ THRESHOLD - Throttle pressure threshold for AI enrichment delay.
- GSS_
WRAP_ OVERHEAD_ BYTES - GSS wrap token overhead per compound RPC (RFC 4121 S4.2.6.2).
- HOT_
DIR_ CUTOFF_ SECS - Age cutoff for orphaned
.tmp.*files during cleanup (seconds). Files older than this are considered abandoned and removed. - HYDRATION_
ALLOCATION_ WARN_ INTERVAL_ SECS - Minimum interval between memory allocation warning log messages during hydration.
- HYDRATION_
ALLOC_ RECHECK_ INTERVAL_ SECS - Interval between allocation rechecks in hydration worker loop (seconds).
- HYDRATION_
BACKPRESSURE_ SLEEP_ ITERATIONS - Retry iterations during backpressure stall before aborting scan.
- HYDRATION_
BACKPRESSURE_ SLEEP_ MS - Sleep between backpressure stall retries (milliseconds).
- HYDRATION_
BACKPRESSURE_ TIMEOUT_ ATTEMPTS - Consecutive back-pressure timeouts before hydration gives up on a job.
- HYDRATION_
BATCH_ BUFFER_ CAPACITY - Batch buffer capacity for the hydration manager event loop.
- HYDRATION_
BATCH_ LIMITS_ HDD - Hydration batch limits for HDD targets: (min_batch, max_batch).
- HYDRATION_
BATCH_ LIMITS_ NETWORK - Hydration batch limits for network (NFS) targets: (min_batch, max_batch).
- HYDRATION_
BATCH_ LIMITS_ NVME - Hydration batch limits for NVMe targets: (min_batch, max_batch).
- HYDRATION_
BATCH_ LIMITS_ SDCARD - Hydration batch limits for SD card targets: (min_batch, max_batch).
- HYDRATION_
BATCH_ LIMITS_ SSD - Hydration batch limits for SSD targets: (min_batch, max_batch).
- HYDRATION_
CHECKPOINT_ INTERVAL - Log progress every N directories during hydration scan.
- HYDRATION_
COPY_ BACKOFF_ BASE_ MS - Base delay for exponential backoff on hydration copy retries (ms).
- HYDRATION_
COPY_ BACKOFF_ MAX_ MS - Maximum backoff delay for hydration copy retries (ms).
- HYDRATION_
COPY_ JITTER_ MAX_ MS - Maximum random jitter added to hydration copy backoff (ms).
- HYDRATION_
COPY_ MAX_ ATTEMPTS - Maximum retry attempts for a single hydration copy operation.
- HYDRATION_
DIRTY_ RANGE_ RING_ SIZE - io_uring ring size for dirty-range replay operations.
- HYDRATION_
FALLBACK_ BUFFER_ COUNT - Buffer count when optimal allocation fails (16 x 4MB = 64MB).
- HYDRATION_
FALLBACK_ CHUNK_ SIZE - Fallback chunk size in bytes when optimal allocation fails (4 MiB).
- HYDRATION_
HEAD_ VERIFY_ BYTES - Bytes to read for head-hash content verification in hydration sync_file_needed. 8KB covers the observed corruption pattern (zeros starting at byte 2).
- HYDRATION_
HEARTBEAT_ INTERVAL_ SECS - Interval between hydration worker heartbeat log messages.
- HYDRATION_
HOT_ DIR_ THRESHOLD_ SECS - Directories modified within this window are considered “hot” and scanned first.
- HYDRATION_
HYSTERESIS_ DOWN - Hysteresis multiplier for transitioning from yield -> primary mode.
- HYDRATION_
HYSTERESIS_ UP - Hysteresis multiplier for transitioning from primary -> yield mode.
- HYDRATION_
IOURING_ RING_ PADDING - Extra ring slots added beyond buffers_per_ring for safety margin.
- HYDRATION_
JOB_ RECV_ TIMEOUT_ SECS - Timeout for receiving the next job from the hydration queue.
- HYDRATION_
LOG_ INTERVAL - Log progress every N files during hydration copy.
- HYDRATION_
LOW_ WATERMARK_ SLEEP_ MS - Sleep at low watermark during scan (milliseconds).
- HYDRATION_
MEMORY_ SHARE_ PRIMARY - Fraction of global buffer budget allocated to hydration in primary mode.
- HYDRATION_
MEMORY_ SHARE_ YIELD - Fraction of global buffer budget allocated to hydration in yield mode.
- HYDRATION_
OOM_ RETRY_ DELAY_ SECS - Delay before retrying a hydration copy after an out-of-memory error.
- HYDRATION_
PATH_ RESOLVE_ POLL_ MS - Polling interval when waiting for path resolution during hydration (ms).
- HYDRATION_
PATH_ RESOLVE_ TIMEOUT_ SECS - Timeout for resolving an inode to a filesystem path during hydration.
- HYDRATION_
QUEUE_ CAPACITY - Maximum pending jobs in the hydration (resync) work queue.
- HYDRATION_
RING_ BUFFER_ CAP - io_uring ring size per hydration worker. Caps memory usage (e.g. 32 x 4MB = 128MB per worker instead of unbounded).
- HYDRATION_
SMALL_ FILE_ THRESHOLD - Small file threshold for hydration fast-path (skip io_uring overhead).
- HYDRATION_
STORM_ THRESHOLD_ MULTIPLIER - Multiplier applied to
global_buffer_limitto derive the storm threshold. - HYDRATION_
TIMEOUT_ FLOOR_ DEFAULT - Hydration timeout floors for unknown/default targets: (stall_secs, overall_secs, postcopy_secs).
- HYDRATION_
TIMEOUT_ FLOOR_ HDD - Hydration timeout floors for HDD targets: (stall_secs, overall_secs, postcopy_secs).
- HYDRATION_
TIMEOUT_ FLOOR_ NETWORK - Hydration timeout floors for Network/NFS targets: (stall_secs, overall_secs, postcopy_secs).
- HYDRATION_
TIMEOUT_ FLOOR_ SSD - Hydration timeout floors for SSD targets: (stall_secs, overall_secs, postcopy_secs).
- HYDRATION_
WORKER_ LOG_ INTERVAL - Log worker liveness every N processed jobs.
- HYDRATION_
YIELD_ MODE_ BUFFER_ DIVISOR - Buffer pool capacity divisor when entering yield mode.
- IDENTITY_
EVICTION_ TTL_ SECS - Time-to-live for stale entries in the inode identity map (seconds).
- IDENTITY_
LRU_ SIZE_ DIVISOR - LRU size divisor: lru_size = max_entries / this.
- IDENTITY_
MAP_ MAX_ ENTRIES - Maximum entries in the identity map after clamping.
- IDENTITY_
MAP_ MIN_ ENTRIES - Minimum entries in the identity map after clamping.
- IDENTITY_
SCAN_ SLEEP_ INTERVAL - Sleep every N entries during identity scan for throttling.
- IDENTITY_
SCAN_ SLEEP_ US - Sleep duration during identity scan throttling (microseconds).
- IDENTITY_
SCAN_ YIELD_ INTERVAL - Yield to scheduler every N entries during identity scan.
- IOURING_
COMPLETION_ TIMEOUT_ STREAK - Consecutive io_uring completion timeouts before the worker declares a stall.
- IO_
URING_ READABLE_ TIMEOUT_ MS - Deadline for io_uring completion poll before giving up on a single wait cycle.
- KRB5P_
MAX_ WRITE_ DATA - Maximum file data size for krb5p compound RPC bypass (bytes).
- KRB5_
TICKET_ MIN_ REMAINING_ SECS - Minimum Kerberos ticket remaining lifetime before renewal (seconds).
- LITE_
HASH_ DEFAULT_ THRESHOLD - Default file-size threshold for lite (head+tail) hashing (128 KiB). Files smaller than this get a full BLAKE3 hash; larger files use a head+tail 4 KiB sample for fast change detection.
- LOG_
THROTTLE_ INTERVAL_ SECS - Minimum interval between repeated log messages during io_uring stall detection.
- MAX_
LEAVES_ PER_ XATTR - Max Merkle leaves that fit in a 64KB xattr with fat nodes. (65536 - 56 overhead) / 44 bytes per leaf (32 hash + 4 entropy + 8 simhash) = 1488.
- MAX_
RETRY_ BACKOFF_ SECS - Ceiling for exponential retry backoff across all retry loops.
- MAX_
WORKER_ CORES - Upper bound on worker threads the daemon will spawn.
- MERKLE_
DELTA_ THRESHOLD - Minimum file size for the Merkle delta copy path in hydration.
- MERKLE_
INDEX_ CHUNK_ SIZE - Merkle chunk size for files stored in the per-target index. 512KB gives a practical tradeoff: 640KB index per 10GB file, 64MB per 1TB.
- MERKLE_
INDEX_ THRESHOLD - Files above this threshold use the per-target Merkle index (.foxing_meta/merkle.db) instead of xattr storage, enabling finer-grained chunk sizes.
- MERKLE_
XATTR_ MAX_ BYTES - Maximum serialized Merkle signature size that fits in a Linux xattr.
- MINIMUM_
ALIGNMENT_ BYTES - Minimum I/O alignment for direct/O_DIRECT writes (bytes). Matches typical sector size.
- MIN_
CHUNK_ SIZE - Minimum chunk size for Merkle tree (64KB). Adaptive chunk_size scales up for large files.
- MIN_
HOLE_ PUNCH_ SIZE - Minimum hole size for FALLOC_FL_PUNCH_HOLE operations (16KB).
- MIN_
INODES_ THRESHOLD - Minimum free inodes on the target before the daemon warns about exhaustion.
- MONITOR_
POLL_ INTERVAL_ MS - Remote monitor poll interval for status/metrics fetching (milliseconds).
- MOUNT_
HEALTH_ CHECK_ INTERVAL_ SECS - Interval between target mount health checks in the hydration manager (seconds).
- MOUNT_
PROBE_ FSYNC_ TIMEOUT_ SECS - Timeout for the fsync liveness probe that detects stale NFS mounts.
- NFS_
BATCH_ ADDITIVE_ INCREASE - Additive increase step for BatchTuner AIMD batch size growth.
- NFS_
BATCH_ EWMA_ ALPHA - EWMA smoothing factor for BatchTuner queue depth and RTT tracking.
- NFS_
BATCH_ MULTIPLICATIVE_ DECREASE - Multiplicative decrease factor for BatchTuner AIMD batch size reduction.
- NFS_
BATCH_ PRESSURE_ HIGH - Queue pressure ratio above which BatchTuner increases batch size.
- NFS_
BATCH_ PRESSURE_ LOW - Queue pressure ratio below which BatchTuner decreases batch size.
- NFS_
BATCH_ READ_ HEAD_ FILES_ PER_ COMPOUND - Max files per batch_read_head compound (SEQUENCE+PUTFH+[LOOKUP+READ]×N). Conservative for 8KB data responses (24KB reply per compound).
- NFS_
CA_ MAX_ OPERATIONS - Requested
ca_maxoperationsin CREATE_SESSION. The server negotiates down to its actual limit; this is the client’s opening bid. - NFS_
CLIENT_ POOL_ SIZE - NFS client pool session count.
- NFS_
COMPOUND_ BYTE_ LIMIT - Maximum compound RPC byte budget before splitting into a new compound (800 KiB).
Keeps each compound well within the server’s
ca_maxrequestsize(~16 MiB). - NFS_
COMPOUND_ TUNER_ INITIAL_ MAX_ OPS - Initial
max_files_per_compoundforCompoundSizeTunerbefore AIMD adaptation. - NFS_
DEFAULT_ PORT - Default NFS server port (IANA-assigned for NFSv4).
- NFS_
GRACE_ RETRY_ COUNT - Number of grace period retries for NFS WRITE operations.
- NFS_
HOSTNAME_ BUFFER_ SIZE - Hostname buffer for gethostname() syscall.
- NFS_
KRB5P_ MAX_ BATCH - Maximum files per compound RPC for krb5p transport. Reduced to avoid accumulating large batches that serialize into many single-file compounds wastefully under AES encryption.
- NFS_
MAX_ REQUEST_ SIZE - Maximum NFSv4.2 request size for channel negotiation (16MB data + 4KB header).
- NFS_
MAX_ RESPONSE_ SIZE - Maximum NFSv4.2 response size for channel negotiation (1MB).
- NFS_
NSS_ BUFFER_ SIZE - NSS getpwuid_r / getgrgid_r username/group lookup buffer size.
- NFS_
PRIVILEGED_ PORT_ END - End (exclusive) of the privileged port range for NFS source-port binding.
- NFS_
PRIVILEGED_ PORT_ START - Start of the privileged port range for NFS source-port binding.
NFS servers with
sec=sysmay require connections from ports < 1024. - NFS_
PROBE_ TIMEOUT_ SECS - Timeout for the NFS NULL RPC liveness probe (connect + read + write).
- NFS_
RDMA_ MAX_ BATCH - Maximum files per compound RPC for RDMA transport. RDMA_MAX_INLINE=4188B – multi-file compounds exceed inline capacity.
- NFS_
RDMA_ MAX_ BATCH_ BYTES - Maximum compound byte budget for RDMA transport (bytes). Sized to fit within RDMA_MAX_INLINE (4188B) with header overhead.
- NFS_
RDMA_ PORT - IANA-assigned NFS/RDMA port (RFC 8267).
- NFS_
RECONNECT_ SLEEP_ MS - Sleep duration between retries when the NFS server is in grace period.
- NFS_
REPLY_ INITIAL_ CAPACITY - Initial capacity for the RPC reply buffer
Vec<u8>. Sized to hold a typical small-compound response without reallocation. - NFS_
SENDFILE_ MIN_ SIZE - Minimum file size to activate the NFS sendfile fast-path (64 KiB). Below this, io_uring overhead is negligible and sendfile adds no benefit.
- NFS_
TCP_ CONNECT_ TIMEOUT_ SECS - TCP connect timeout for NFSv4.2 client session recovery and privileged port fallback.
- NFS_
TCP_ MAX_ BATCH - Maximum files per compound RPC for TCP/TLS AUTH_SYS transport.
- NFS_
TCP_ MAX_ BATCH_ BYTES - Maximum compound byte budget for TCP/TLS transport (8 MiB).
- NFS_
TCP_ RW_ TIMEOUT_ SECS - TCP read/write timeout for NFSv4.2 compound RPC streams.
- NFS_
TLS_ ALPN - ALPN protocol identifier for RPC-over-TLS (RFC 9289 S7.2).
- NFS_
TLS_ HANDSHAKE_ TIMEOUT_ SECS - Timeout for the TLS handshake after AUTH_TLS STARTTLS upgrade (seconds).
- NFS_
WRITER_ CHANNEL_ CAPACITY - Channel capacity for the NFS writer task’s job queue.
Workers submit
NfsWriteJobentries; the writer drains in batch cycles. - NFS_
XDR_ ENCODER_ BODY_ CAPACITY - Initial XDR encoder capacity for compound RPC body.
- NFS_
XDR_ ENCODER_ SMALL_ CAPACITY - Initial XDR encoder capacity for small RPC messages (EXCHANGE_ID, CREATE_SESSION).
- NFS_
XDR_ ENCODER_ TINY_ CAPACITY - Initial XDR encoder capacity for GSS RPC header.
- NFS_
XDR_ HEADER_ OVERHEAD - Estimated XDR header overhead per compound RPC (bytes). Covers SEQUENCE, PUTFH root, and framing. Used as the initial byte accumulator.
- NFS_
XDR_ PER_ OP_ OVERHEAD - Estimated per-operation XDR overhead (bytes). Covers PUTFH + OPEN + WRITE + CLOSE op headers and filehandle encoding for each file in a compound.
- ONE_
SHOT_ CHECK_ INTERVAL_ MS - One-shot sync mode hydration completion check interval (milliseconds).
- OUTAGE_
FULL_ SCAN_ THRESHOLD_ SECS - Outage duration (seconds) beyond which recovery triggers a full scan instead of journal replay.
- OUTAGE_
JOURNAL_ MAX_ ENTRIES - Maximum events stored in the outage journal before oldest entries are dropped.
- POSTCOPY_
TIMEOUT_ SECS Deprecated - PROCESS_
SEGMENT_ STALL_ SECS Deprecated - PROCESS_
SEGMENT_ TIMEOUT_ SECS Deprecated - PROFILE_
HDD_ VALUES - I/O profile for HDD targets: (min_chunk, max_chunk, min_qd, max_qd).
- PROFILE_
NETWORK_ VALUES - I/O profile for network (NFS) targets: (min_chunk, max_chunk, min_qd, max_qd).
- PROFILE_
NVME_ VALUES - I/O profile for NVMe targets: (min_chunk, max_chunk, min_qd, max_qd).
- PROFILE_
SDCARD_ VALUES - I/O profile for SD card targets: (min_chunk, max_chunk, min_qd, max_qd).
- PROFILE_
SSD_ VALUES - I/O profile for SSD targets: (min_chunk, max_chunk, min_qd, max_qd).
- PROGRESS_
INTERVAL_ HUMAN_ MS - Progress reporting interval in human-readable output mode (milliseconds).
- PROGRESS_
INTERVAL_ JSON_ MS - Progress reporting interval in JSON output mode (milliseconds).
- QCOW2_
BITMAPS_ EXTENSION - Header extension type for persistent dirty bitmaps.
- QCOW2_
BITMAP_ FLAG_ AUTO - Bitmap flag: auto (actively recording writes).
- QCOW2_
BITMAP_ FLAG_ IN_ USE - Bitmap flag: in_use (inconsistent – was not properly synced).
- QCOW2_
BITMAP_ TYPE_ DIRTY - Dirty tracking bitmap type identifier.
- QCOW2_
HEADER_ V3_ SIZE - qcow2 v3 header size (minimum).
- QCOW2_
L2_ OFFSET_ MASK - L1/L2 host offset extraction mask (bits 9-55).
- QCOW2_
MAGIC - qcow2 magic bytes: “QFI\xfb” in big-endian.
- QCOW2_
MAX_ BACKING_ PATH - Maximum backing file path length (qcow2 spec).
- RAYON_
HASH_ THRESHOLD - Minimum file size for parallel BLAKE3 hashing via
update_rayon()(1 MiB). Below this threshold, single-threaded hashing is faster due to rayon overhead. - RDMA_
CQ_ DEPTH - Completion queue depth for RDMA ibverbs CQ creation.
- RDMA_
CQ_ POLL_ MAX_ SPINS - Maximum CQ poll spins before declaring an RPC timeout. At ~10ns per spin, 10M spins ~= 100ms busy-wait ceiling.
- RDMA_
INLINE_ PADDING - Padding/overhead added to inline body for RDMA message framing (bytes).
- RDMA_
MAX_ INLINE_ BODY - Maximum inline NFS body size for RDMA Send/Recv operations (bytes).
Must fit within
RDMA_MAX_INLINE= body + header + padding. - RDMA_
RESOLVE_ TIMEOUT_ MS - Timeout for RDMA CM address/route resolution (milliseconds).
- RDMA_
RPC_ TIMEOUT_ SECS - Fallback RPC timeout for RDMA operations (seconds).
Used when CQ polling exhausts
RDMA_CQ_POLL_MAX_SPINSwithout completion. - RECOVERY_
SCAN_ TIMEOUT_ SECS - Timeout for a targeted recovery scan to complete (seconds).
- REORDER_
BUFFER_ BYTES - Maximum memory budget for the reorder buffer (bytes).
- REORDER_
BUFFER_ CAPACITY - Maximum number of events the reorder buffer can hold before dropping.
- REORDER_
PENDING_ PANIC_ COUNT - Pending event count that triggers panic-mode forced delivery in ReorderBuffer.
- REORDER_
PENDING_ WARN_ COUNT - Pending event count that triggers warning-level stall timeout in ReorderBuffer.
- REORDER_
PRESSURE_ PANIC_ PCT - Reorder buffer fill fraction that triggers panic-mode forced delivery.
- REORDER_
PRESSURE_ STALL_ TIMEOUT_ MS - Stall timeout under pressure conditions in ReorderBuffer (milliseconds).
- REORDER_
PRESSURE_ WARN_ PCT - Reorder buffer fill fraction that triggers a warning log.
- REORDER_
SOFT_ TIMEOUT_ MAX_ MS - Maximum soft timeout for the reorder buffer (ms). Caps wait time for missing sequences.
- REORDER_
SOFT_ TIMEOUT_ MIN_ MS - Minimum soft timeout for the reorder buffer before releasing out-of-order events (ms).
- REORDER_
ZOMBIE_ TIMEOUT_ MS - Time after which an undelivered reorder buffer entry is considered a zombie (ms).
- RETRY_
QUEUE_ STALL_ THRESHOLD_ ITERATIONS - Retry queue iterations with zero progress before triggering a forced drain.
- SECTOR_
SIZE_ BYTES - Standard disk sector size (bytes). Used for sparse file detection
(
blocks * SECTOR_SIZE < file_size / 2) and I/O alignment fallback. - SENDFILE_
CHUNK_ SIZE - Per-call chunk size for
sendfile(2)in the copy-file helper (2 MiB). Prevents a single sendfile from monopolizing the kernel for too long. - SEQUENTIAL_
PROXIMITY_ THRESHOLD_ BYTES - Sequential I/O proximity threshold (bytes). I/O within this range of the last sequential offset is treated as sequential for scheduling purposes.
- SHUTDOWN_
DEADLINE_ SECS - Grace period for worker tasks to complete during daemon shutdown (seconds).
- SIDECAR_
PREFIX_ BUDGET - Character budget for the sidecar type prefix in an enrichment prompt.
- SMALL_
FILE_ THRESHOLD - File size boundary for copy strategy selection (64 KiB). Files smaller than this use sendfile; larger files use io_uring pipeline.
- SMB_
SUPER_ MAGIC - SMB/CIFS filesystem magic number (from statfs).
- SPARSE_
HOLE_ RATIO_ SKIP_ THRESHOLD - When a file’s hole ratio (bytes in holes / total bytes) exceeds this threshold, skip fallocate pre-allocation. The io_uring pipeline will punch holes as needed. This avoids the wasteful allocate-then-punch round-trip for borderline-sparse files.
- SSE_
KEEPALIVE_ INTERVAL_ SECS - Interval between SSE keepalive comments to prevent proxy/client timeouts (seconds).
- SSE_
POLL_ INTERVAL_ MS - SSE event stream poll sleep when no new events are available (milliseconds).
- SSE_
PULL_ BATCH_ SIZE - Maximum entries pulled from the EventBus per SSE poll cycle.
- STALL_
WATCHDOG_ INTERVAL_ SECS - How often the stall watchdog checks for stuck workers.
- STALL_
WATCHDOG_ THRESHOLD_ SECS - Seconds without progress before the watchdog declares a worker stalled.
- STDIN_
CHUNK_ SIZE - Read chunk size for stdin pipe mode (1 MiB). Balances memory usage against syscall frequency for piped input.
- STORM_
THRESHOLD_ MAX - Maximum rename storm threshold before forced drain.
- STORM_
THRESHOLD_ MIN - Storm detection threshold range for event burst detection.
- STRATIS_
MONITOR_ INTERVAL_ SECS - Interval between Stratis pool capacity checks via D-Bus (seconds).
- STRATIS_
POOL_ CRITICAL_ PCT - Stratis pool free-space percentage that triggers critical warnings and emergency version pruning to reclaim capacity.
- STRATIS_
POOL_ LOW_ PCT - Stratis pool free-space percentage that triggers low-space warnings.
- SYNC_
MIN_ FREE_ BYTES - Minimum free bytes on target before fxcp sync aborts with an error (10 MiB).
- SYSTEM_
GOVERNOR_ UPDATE_ INTERVAL_ SECS - SystemGovernor periodic update interval (seconds).
- SYSTEM_
MEMORY_ CONSTRAINED_ MB - Systems with less than this are considered memory-constrained.
- SYSTEM_
MEMORY_ LARGE_ MB - Systems with more than this get expanded buffers.
- TARGET_
ERROR_ STREAK_ THRESHOLD - Consecutive target errors before the worker requests a rescan on recovery.
- TLSH_
ENTROPY_ SKIP_ THRESHOLD - Entropy threshold above which TLSH is skipped (encrypted/compressed data).
- TLSH_
MIN_ DATA_ SIZE - Minimum data length in bytes for TLSH computation.
- TUI_
BANDWIDTH_ HISTORY_ CAPACITY - Bandwidth history ring buffer capacity for TUI sparkline.
- TUI_
BUFFER_ UTIL_ HISTORY_ CAPACITY - Buffer utilization history capacity for TUI sparkline.
- TUI_
CHANNEL_ CAPACITY - Channel capacity for TUI event and command channels.
- TUI_
LATENCY_ HISTORY_ CAPACITY - Latency history ring buffer capacity for TUI sparkline.
- TUI_
LOG_ BUFFER_ CAPACITY - Log buffer capacity for TUI log panel.
- TUI_
TICK_ INTERVAL_ MS - TUI tick/refresh interval (milliseconds).
- TUNER_
AIMD_ ADDITIVE_ STEP - AIMD additive increase step for batch size growth.
- TUNER_
AIMD_ MULTIPLICATIVE_ FACTOR - AIMD multiplicative decrease factor on congestion detection.
- TUNER_
BDP_ PACING_ GAIN_ DRAIN - BBR pacing gain during drain phase (0.5x = half rate to clear queues).
- TUNER_
BDP_ PACING_ GAIN_ MUTED - BBR pacing gain when muted by system pressure (0.75x = reduced throughput).
- TUNER_
BDP_ PACING_ GAIN_ PROBE - BBR pacing gain during bandwidth probing (1.25x = slight overshoot).
- TUNER_
BDP_ PACING_ GAIN_ STARTUP - BBR pacing gain during startup phase (2x = double estimated BDP).
- TUNER_
DEFAULT_ AVG_ EVENT_ SIZE_ BYTES - Default assumed average event payload size (bytes).
- TUNER_
DEFAULT_ BANDWIDTH_ GB_ PER_ SEC - Default estimated bandwidth for new targets before probing (GB/s).
- TUNER_
DEFAULT_ BATCH_ SIZE - Default batch size for worker event processing.
- TUNER_
DEFAULT_ COALESCE_ BYTES - Default coalesce threshold for merging small writes (bytes).
- TUNER_
DEFAULT_ FLUSH_ INTERVAL_ US - Default flush interval for worker event batches (microseconds).
- TUNER_
DEFAULT_ FLUSH_ MAX_ US - Maximum flush interval ceiling (microseconds).
- TUNER_
DEFAULT_ FLUSH_ MIN_ US - Minimum flush interval floor (microseconds).
- TUNER_
DEFAULT_ LATENCY_ SPIKE_ THRESHOLD_ US - Latency spike detection threshold (microseconds); exceeding triggers drain.
- TUNER_
DEFAULT_ MEMORY_ PRESSURE_ PCT - Memory pressure percentage threshold for tuner v2 throttling.
- TUNER_
DEFAULT_ PENDING_ QUEUE_ MAX - Maximum pending queue depth before events are dropped.
- TUNER_
DEFAULT_ PENDING_ QUEUE_ MIN - Minimum pending queue depth before back-pressure engages.
- TUNER_
DEFAULT_ QUEUE_ DEPTH - Default io_uring submission queue depth.
- TUNER_
DEFAULT_ QUEUE_ DEPTH_ ALPHA - EWMA smoothing factor for queue depth estimation.
- TUNER_
DRAIN_ TIMEOUT_ SECS - Maximum time the tuner stays in drain phase before returning to steady state.
- TUNER_
EWMA_ ALPHA - EWMA smoothing factor for tuner metrics (lower = smoother, slower to react).
- TUNER_
HIGH_ BANDWIDTH_ THRESHOLD_ GB_ PER_ SEC - Bandwidth threshold above which aggressive probing is used (GB/s).
- TUNER_
HYDRATION_ DEBOUNCE_ US - Debounce window for hydration tuner updates (microseconds).
- TUNER_
IDLE_ TIMEOUT_ SECS - Seconds of zero throughput before the tuner enters idle mode.
- TUNER_
MAX_ BATCH_ SIZE - Hard upper bound on batch size regardless of tuner state.
- TUNER_
MAX_ COALESCE_ BYTES - Maximum coalesce threshold (4MB) – memory guardrail.
- TUNER_
MAX_ FLUSH_ US - Upper bound on flush interval (microseconds). Caps how long writes can be batched.
- TUNER_
MEMORY_ PRESSURE_ THRESHOLD - Memory usage fraction that triggers tuner throttling.
- TUNER_
MIN_ BATCH_ SIZE - Hard lower bound on batch size to prevent starvation.
- TUNER_
MIN_ FLUSH_ US - Lower bound on flush interval (microseconds). Prevents excessive flushing.
- TUNER_
STARTUP_ DURATION_ SECS - Duration of the BBR startup phase (aggressive probing).
- TUNER_
STATE_ TRANSITION_ DELAY_ US - Minimum delay between BBR tuner state transitions (microseconds).
- TUNER_
V2_ EWMA_ ALPHA_ DEFAULT - Default EWMA alpha for tuner v2 smoothing filters.
- TUNER_
WINDOW_ SECS - Sliding window duration for tuner throughput and latency sampling.
- TUNING_
BOARD_ CONSENSUS_ INTERVAL_ SECS - TuningBoardManager global consensus interval (seconds).
- TUNING_
BOARD_ MANAGER_ CAPACITY - Maximum targets tracked by TuningBoardManager.
- VERSION
- Crate version string, pulled from Cargo.toml at compile time.
- VERSION_
LIST_ DISPLAY_ LIMIT - Maximum versions shown in
snapshot listoutput. - VERSION_
SCAN_ POLL_ INTERVAL_ MS - Poll sleep interval for version index scan (milliseconds).
- WAL_
BATCH_ TOKENS_ INITIAL - Initial token count for WAL batch admission control. Each write operation consumes a token; tokens regenerate on flush.
- WAL_
MAX_ WRITERS_ BEFORE_ BARRIER - Maximum concurrent writers before the WAL engages barrier mode. Beyond this threshold, new writers block until the current batch completes.
- WAL_
SEQUENCE_ OVERFLOW_ THRESHOLD - Sequence number proximity to
u64::MAXthat triggers an overflow reset. Whenglobal_seq >= u64::MAX - threshold, the sequence resets to 1. - WORKER_
BUFFER_ POOL_ ALLOC_ ATTEMPTS - Number of attempts to allocate a buffer from the worker pool before giving up.
- WORKER_
CLEANUP_ INTERVAL_ SECS - Interval between worker housekeeping passes (stale entry cleanup, stats flush).
- WORKER_
FLUSH_ CAP_ PER_ TICK - Maximum events drained from the coalescer per flush tick. Prevents unbounded processing in a single select iteration.
- WORKER_
INITIAL_ IO_ LATENCY_ SECS - Grace period after worker startup before I/O latency measurements are trusted.
- WORKER_
IO_ URING_ DEPTH_ MIN - Minimum io_uring submission queue depth per worker.
- WORKER_
METRICS_ UPDATE_ INTERVAL_ MS - How often workers flush Prometheus metric gauges (ms).
- WORKER_
MOUNT_ CHECK_ INTERVAL_ MS - How often workers check whether the target mount is still alive (ms).
- WORKER_
REORDER_ BUFFER_ MIN_ BYTES - Minimum reorder buffer size before the worker starts processing events (bytes).
- WORKER_
RETRY_ QUEUE_ BASE_ BACKOFF_ MS - Base backoff delay for the worker retry queue (ms). Doubles on each attempt.
- WORKER_
RETRY_ QUEUE_ MAX_ RETRIES - Maximum retries for a single event in the worker retry queue.
- WORKER_
STARTUP_ CHECK_ DELAY_ MS - Delay after worker startup before running the first health check (ms).
- WORKER_
TUNE_ EVENT_ THRESHOLD - Minimum events processed before the tuner considers adjusting parameters.
- WORKER_
TUNE_ INTERVAL_ US - Interval between adaptive tuner recalculations (microseconds).
- XFS_
IOC_ EXCHANGE_ RANGE - XFS_IOC_EXCHANGE_RANGE ioctl for atomic file content exchange.
Statics§
- ONE_
SHOT_ MODE - When true, the daemon runs a single sync pass and exits (fxcp mode).