v1.4.0

New Features

• Added HTTP/SSE server replacing the previous WebSocket service. The server
  subcommand now starts an HTTP server with REST endpoints and SSE search
  streaming. Key additions:
  • Three MVP endpoints: GET /health, GET /api/v1/system/info,
    POST /api/v1/search/stream (SSE streaming search with progress, element
    batches, cancellation on disconnect, and max-results limit)
  • Parallelized SSE search with bounded mpsc channel (capacity 32)
    backpressure; element batches are never dropped, progress events may be
    coalesced
  • Terminal event invariant: exactly one of completed, cancelled, or
    error. All three carry a SearchStreamResult payload with matched
    elements, source-file counts, compressed-byte metadata, rate, and matching
    collectors/files
  • Server admission control (server_max_concurrent_searches, default 3);
    excess requests return HTTP 429. SSE search returns 400 on invalid
    peer_ip values
  • Phase 2 REST API: stateless endpoints (/time/parse, /country/lookup,
    /ip/lookup, GET /ip/public), DB read-only endpoints
    (/database/status, /rpki/roa/lookup, /rpki/aspa/lookup,
    /pfx2as/lookup, /as2rel/relationship, POST /as2rel/search),
    DB refresh endpoints (/database/refresh, /inspect/refresh,
    /as2rel/refresh), and composite endpoints (/rpki/roa/validate,
    /inspect/query). ASPA validation is deferred. DB-backed endpoints
    return NOT_INITIALIZED (HTTP 503) if required data is missing
  • Token-based auth middleware (server_auth_enabled / server_auth_token);
    when enabled, /api/v1/* requires Authorization: Bearer <token>
    (case-insensitive per RFC 7235); /health stays open. Server refuses to
    start if auth is enabled but token is empty
  • CLI remote search mode: --remote-url flag sends queries to a remote
    Monocle HTTP service; --remote-token provides Bearer auth. Consumes SSE
    stream and formats results with existing CLI formatters. Remote search
    exits with non-zero status on error/cancelled SSE events or connection
    drop without completed. Applies --filter-file/--prefix-file merging
    and filter validation before dispatch
  • Docker Compose deployment with health check, persistent volumes, and
    env-based configuration. Docker runtime runs as non-root monocle user
    with pre-created config directory and HOME set
  • Server configuration via monocle.toml and MONOCLE_* environment
    variables: server_address, server_port, server_max_search_batch_size,
    server_max_search_results, server_search_timeout_secs,
    server_max_concurrent_searches, search_concurrency
• Added Cisco sh ip bgp text dump parsing to the parse command. The parser
  auto-detects the fixed-width plaintext format from the preamble, extracts
  prefix/AS-path/next-hop/metric/local-pref/origin, and supports all standard
  parse filters, output formats, and MRT export (--mrt-type rib|updates).
  Multi-path continuation lines and wrapped-prefix layouts are handled.
  Header column detection uses whitespace split rather than fixed positions.
• Added --filter-file (JSON) and --prefix-file (newline text) flags to
  monocle parse and monocle search for loading large filter sets from
  files. File filters merge with CLI flags — union within each dimension (OR),
  AND across dimensions. Supports the RIB-extract → filter-updates workflow at
  scale (#117).
• Added RPKISPOOL as the default historical RPKI source with Sobornost as the
  default mirror. The rpki roas and rpki aspas commands retain ripe and
  rpkiviews; invalid source/collector combinations now return errors (#134).
• Added --use-cache / --cache-dir MRT-file caching and --fields output
  selection to the rib command (#137).
• Added --concurrency to local search and server commands. Explicit values
  use local rayon thread pools; unset/0 keeps rayon defaults including
  RAYON_NUM_THREADS.

Breaking Changes

• Replaced WebSocket server with HTTP/SSE service. All WebSocket modules
  (protocol, handler, sink, op_sink, router, operations,
  handlers/*) have been removed.
• Updated axum to 0.8 and tower-http to 0.6.
• Removed uuid and async-trait dependencies (no longer needed without
  WebSocket operation tracking).
• Updated ServerArgs CLI flags: removed --data-dir, --max-concurrent-ops,
  --max-message-size, --connection-timeout-secs, --ping-interval-secs;
  added --max-search-batch-size, --max-search-results,
  --search-timeout-secs. --address and --port are now optional (default
  to config values).

Performance Improvements

• Optimized database refresh (bulk insert) performance by 21–49% across all
  repositories. Indexes are now dropped before bulk insert and rebuilt in one
  pass afterward, instead of being updated per-row.
• Removed redundant idx_pfx2as_prefix_str and low-value
  idx_pfx2as_validation indexes. lookup_exact rewritten to use the
  existing BLOB range index (prefix_start/prefix_end/prefix_length).
• Fixed PRAGMA restore bug: store() methods no longer toggle
  synchronous/journal_mode PRAGMAs, preserving the connection's
  WAL/NORMAL defaults. Previously, every refresh left the connection
  in DELETE/FULL mode, degrading query performance until restart.
• Switched INSERT OR REPLACE to plain INSERT in all refresh paths
  (tables are cleared before insert, so no conflicts are possible).
• Added db_refresh_bench example for measuring refresh performance
  with real or synthetic data.
• Added tracing::info! progress logging to the rib command (visible with
  --debug): RIB download/parse reports message counts, and update replay
  shows per-file progress.

Bug Fixes

• Fixed search --dump-type rib applying the requested snapshot time window to
  per-route timestamps inside matching RIB dumps. Stable routes learned before
  the dump timestamp are now retained, restoring AS-path search results (#136,
  #138).
• Fixed rib command dropping RIB entries whose per-route timestamp predates
  the RIB dump time. The base RIB parser was incorrectly applying a start_ts
  filter at the dump timestamp, silently excluding stable routes learned days
  or weeks earlier (#124).
• Fixed rib command using an earlier RIB than necessary when the target
  timestamp coincides with a RIB dump time. The broker query's ts_end filter
  is exclusive, so a RIB starting exactly at the target was excluded, forcing
  the code to select the previous RIB and replay unnecessary update files.
  This caused ~3× slowdown for common midnight RIB queries.
• Fixed --time-format rfc3339 being ignored for json, json-line, and
  json-pretty output formats. JSON output now honors --time-format: unix
  (default) emits a numeric timestamp field (backward compatible); rfc3339
  emits an RFC 3339 string (#123).
• Validated prefix input in pfx2as_lookup and roa_lookup REST endpoints
  before spawning blocking tasks, so invalid prefixes return 400 instead of
  500.
• Auth middleware now accepts case-insensitive Bearer scheme tokens per
  RFC 7235.
• POST /api/v1/database/refresh with source=pfx2as now returns HTTP 501
  instead of 200, so automation can detect the operation is not implemented.
• GET /api/v1/rpki/roa/lookup now applies AND semantics when both prefix
  and asn are provided (filters covering ROAs by origin ASN).

Build & Infrastructure

• Replaced QEMU-emulated multi-platform Docker build with native per-platform
  builds: linux/amd64 on ubuntu-latest, linux/arm64 on ubuntu-24.04-arm
  native runners. Each build pushes by digest; a final manifest-merge job
  assembles the multi-architecture OCI image index. Expected to reduce CI build
  time from ~57 minutes to roughly the slower platform's native compile time.
• Added cargo-chef to the Dockerfile for dependency-layer caching. Source-only
  changes now reuse the pre-compiled dependency layer instead of recompiling all
  dependencies from scratch.
• Switched Docker BuildKit cache from GitHub Actions cache to registry-based
  cache (bgpkit/monocle:buildcache-{arch}), removing the 10 GB GHA cache
  budget constraint.

Code Improvements

• Added HistoricalRpkiCollectorOption as the preferred general-purpose alias
  while retaining RpkiViewsCollectorOption compatibility, and clarified
  historical source/collector CLI errors.
• Removed redundant formatting references in parse command output to satisfy
  current Clippy checks.
• Made refresh index rebuilds atomic by wrapping index drops, table clears,
  inserts, and index recreation in one transaction; indexes are dropped before
  clearing tables to avoid unnecessary per-row index maintenance during
  refresh.
• Ensured upgraded pfx2as databases drop removed legacy indexes during
  refresh and kept the benchmark example buildable with
  --no-default-features --features lib.
• Preserved immediate transaction semantics for pfx2as refreshes and reused
  shared AS2Rel schema constants when rebuilding indexes.

Dependencies

• Updated bgpkit-parser to v0.18.0.
  • Default parse JSON output and SSE search element batches now include
    peer_bgp_id for TableDumpV2 RIB entries when the peer's BGP Identifier is
    available.
  • Known-but-unsupported BGP attributes are retained in the existing
    unknown/deprecated fields rather than silently dropped, and malformed
    BGP input is validated more strictly.