- queryNmsAlarms no longer silently clamps page to 2; add is_cleared.
- Bad integer args return invalid_argument tool errors instead of JSON-RPC crashes.
- Nested {"ok": true, "data": {"ok": false}} results are flagged isError.
- stdio: answer ping, never respond to notifications, version from metadata.
- getNmsNe resolves exact host_name/IP; severity lowercased.
- Outputs use nms_ne_id/source=nms; drop ME{uuid} names, echoed field lists, idle hop_* fields.
- Fabric paths: parallel links count as one hop (edges[].parallel_links).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pandas + numpy were ~60MB of the shipped venv for three call sites (alarm
Excel import, managed-NE import, import template). netx_api/tabular.py
reads/writes xlsx and csv and parses spreadsheet timestamps.
Checked against the pandas version on the same inputs: timestamp parsing
identical for 24 formats, managed-NE csv/xlsx import, stored rows and both
templates identical. Differences, all fixes:
- alarm codes in mixed text/number columns are "1001", not "1001.0"
- error row numbers are the real sheet rows (blank rows no longer shift them)
- GBK-encoded CSVs (Chinese Excel "Save as CSV") import instead of failing
- "2024年1月2日 03:04:05" style timestamps parse
Also: conftest compares parsed DB URLs (newer SQLAlchemy renders "C:" as
"C%3A").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Per-user layout lives in the new app_user.prefs JSON column, read via
/v1/auth/me and written via PUT /v1/auth/me/prefs/workbench. The
workbench gets an in-place edit mode (gear -> drag to reorder, x to
remove, tray to add back, reset/done). Reset stores empty prefs so
registry changes still apply; ids of modules the user can't currently
see are kept on save.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add column filters and clearer API error mapping for NE management, tighten port validation, and slim the workbench into reusable modules with cleaner styles.
Co-authored-by: Cursor <cursoragent@cursor.com>
Replace top filter bars with shared Excel-style column menus, fix select-all and modal interactions, and use plain dropdown indicators. Apply exact-value filters before pagination and export. Isolate pytest database state and cover filter behavior.
Save path: one atomic POST /views/{id}/commit (edges, ports, styles, node
add/remove, positions) replaces the multi-request save. Edge styles and
manual-edge port edits now go through dirty/undo/save instead of writing
immediately; non-manual edge ports are read-only.
Data safety: UI saves no longer clear MCP node locks or freeze the device
name into the placement label (label_override + omit-means-keep semantics,
also applied to the MCP positions tool). Adding NEs/neighbors with unsaved
edits now saves first instead of silently dropping them. Positions no longer
bypass membership caps, except persisting synthesized child-region icons.
Fixes found in acceptance: nodes added in the same commit kept (40,40)
under autoflush=False; edits on edges deleted in the same save 404'd the
whole save; box-selection drags and arrow-key moves were not marked dirty;
large canvases far from the origin opened blank and locate did nothing
because fitView waits for measurement that onlyRenderVisibleElements never
does (now fitBounds from known geometry).
Also: batch N+1 lookups in graph/positions, deterministic truncation,
outside-peer scan excludes internal edges, split TopologyPage into hooks,
fix flaky UME/LLDP port test, add scripts/topology_clear_frozen_labels.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Correctness (A): normalize port-map names with the template rules (snapshot v2,
v1 replays unchanged); protect mappings/templates that are referenced; stop
dropping edited mapping text; accept pulled-but-not-shutdown ports as migrated;
validate baseline ownership on project create; one canonical key form; fix
built-in template tolerances and upgrade old packaged values in place.
Cross-vendor identity (B): single identity pipeline for every path, per-side and
per-vendor interface rules, value maps, filters after normalization, mapping
pre-flight for compare and cutover projects, SQL eligibility fixes.
Data quality (V): no_data / suspect_parse / not_collected sheet states, command
slice coverage, directional compare modes, value-sorted duplicate pairing,
before-side raw names in results and CSV.
Robustness (H): bounded auto-run history, background evaluate/finish with
polling, manual route tables out of the acceptance gate, baseline row cache,
scoped duplicate-key failures, merged HF tasks with a per-NE session cap,
same-device cutover, explicit scope_by_port_map, read-only GETs, resolvable red
tickets.
Frontend: pre-flight report, mapping table view, template editor support for the
new rules, data-state banners. Docs: canonical field contract.
Not included: Huawei status parsers (no real samples), SQL-path route compare.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Performance: event-driven stdout pump (no polling thread per socket, no 200ms
coalescing stall), keystrokes written immediately, blocking SSH reads, bounded
deque queue with batch drain, incremental UTF-8 decode, buffered session log,
non-blocking session teardown, shorter login/settle waits, xterm WebGL renderer
and idle-time direct writes. Zero-latency fake device: echo p50 11ms -> 2.3ms,
command RTT 219ms -> 20ms, bulk output 2.1 -> 8.3 MB/s.
Audit: track the device's own input row server-side (webcrt_cmd_audit) and
resolve each Enter from the echoed line feed, so Tab, backspace, cursor-key
edits, history recall, paste, Ctrl-C and pagers record what actually ran.
Password lines are redacted; pending Enters are flushed on close.
Hop: stop closing the hop session when the target prints the same prompt as the
hop (system-view); recognise the Huawei "closed by the remote host" message.
Generic NEs that show a network prompt now get the network key map.
Adds bench_webcrt.py, fake_ssh_device.py and e2e_webcrt_input_scenarios.py.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
With a few fails among tens of thousands of rows, 1-decimal rounding hid residual failures; keep extra precision so the UI never shows 100% when fails remain.
Co-authored-by: Cursor <cursoragent@cursor.com>
Show pending check items instead of a fake 100% pass, publish sheet results as each finishes, and use keyset chunk loads with rows-loaded progress.
Co-authored-by: Cursor <cursoragent@cursor.com>
Load metric rows in LIMIT/OFFSET chunks so large-table compares no longer hit the SQLAlchemy yield_per/unique conflict.
Co-authored-by: Cursor <cursoragent@cursor.com>
Run compares in the background with progress, persist fails first, and store success as sampled key+row_id hydrated from source so million-row sheets no longer block the API.
Co-authored-by: Cursor <cursoragent@cursor.com>
Choose bundled or external PostgreSQL in the Setup wizard with connection
validation; optional NETX_CREDENTIAL_SECRET_KEY; single desktop shortcut;
faster uninstall and tray startup for offline Windows packages.
Co-authored-by: Cursor <cursoragent@cursor.com>
Ship installable releases via zip/Inno Setup with optional bundled Postgres, API-hosted SPA, manual update path, and CI workflow for future tag builds.
Co-authored-by: Cursor <cursoragent@cursor.com>
Expose recommended_mode/hints on getManagedNe, support background exec-jobs with getNeExecJob in MCP, and allow linux_shell on MikroTik as well as Linux.
Co-authored-by: Cursor <cursoragent@cursor.com>
Stop inheriting `(default for vrf)` onto RD-only blocks, and normalize CE vrf commands so `(afi,vrf)` sheet splits no longer collide with global VPNx tables.
Co-authored-by: Cursor <cursoragent@cursor.com>
Create tasks via pick-NE or import log; re-import appends batches like collects, without inventory NE or SSH.
Co-authored-by: Cursor <cursoragent@cursor.com>
List failing profile/command/reason in batch.message (including unmatched and expand_all), and require whitespace/EOL after HH:MM:SS so prefixes like 56:16:10::/64 are parsed.
Co-authored-by: Cursor <cursoragent@cursor.com>
Write JSONL iteratively, flush via chunked PG execute_values, skip mega raw in DB, and yield bgp_route rows instead of building a full list.
Co-authored-by: Cursor <cursoragent@cursor.com>
A defaulted field was inserted before required textfsm_command/vendor fields and broke collect startup.
Co-authored-by: Cursor <cursoragent@cursor.com>
Skip TextFSM for neighbor route tables (status vs plain, From, RD+VRF, heavy wraps); parse via callback and cover with live RR/VRF fixtures.
Co-authored-by: Cursor <cursoragent@cursor.com>
RR vpnv6 wrap lines with next-hop+metrics were collapsing ECMP legs; store raw_line_count/declared_total before DB truncate so collect stats stay accurate.
Co-authored-by: Cursor <cursoragent@cursor.com>
Show collecting over paused schedule, enqueue sync so collect_running flips immediately, and expose export on metric sheets with a card layout for Commands.
Co-authored-by: Cursor <cursoragent@cursor.com>
Decouple heavy CLI parse via ParsePool, keep ECMP/multi-RD keys, and expose per-command CLI lines vs DB rows with raw.txt export.
Co-authored-by: Cursor <cursoragent@cursor.com>
IPv4/IPv6 neighbor in/out keep AF peers and peer-group expand, while top-level global activate is routed to ipv4 or ipv6 discover by address literal.
Co-authored-by: Cursor <cursoragent@cursor.com>
Stamp local_as on peer/route/evpn results and discover (local_as, neighbor/vrf) pairs from config_bgp_peer so each router bgp process is collected separately.
Co-authored-by: Cursor <cursoragent@cursor.com>
Avoid idle Session disconnect after heavy timeout, and separate SSH collect from batched Postgres persist via disk spool.
Co-authored-by: Cursor <cursoragent@cursor.com>
Clear stuck collect_running without re-dispatching; finalize in-flight commands as failed with interrupted_by_restart.
Co-authored-by: Cursor <cursoragent@cursor.com>
Catalog previously only had VPNv4/VPNv6 route items; global ipv4/ipv6 discover requires empty vrf so VRF CE peers are not mixed in.
Co-authored-by: Cursor <cursoragent@cursor.com>
