A 404 from the registry means the image name is wrong or the registry
needs authentication, but the pull wrapped every error with "please
check your Internet connection". Say what actually failed and keep the
network advice for the other errors.
A pull now reports docker.image.pull notifications on the controller
stream: started, throttled downloading events with the layer counts
and the latest daemon message, then completed or failed. An optional
progress_id is echoed in every event so a client can correlate them
with its own request. Image operations belong to a compute, not to a
project, so the events go to the controller stream and reach clients
that have no project open.
The pull also has an idle socket read timeout of 60 seconds. Without
it an unreachable registry left the pull, the compute request and the
client request hanging forever with no feedback at all. The timeout
is idle time, not total time: the daemon streams progress during the
whole pull, so a minute without data means the registry is gone. When
it fires, the daemon connection closes and Docker cancels the pull.
Docker.http_query now accepts an aiohttp ClientTimeout object so a
caller can ask for a policy finer than a total timeout.
The image inventory scanned every file under the image roots and
reported anything without an image header as an invalid image. Files
users legitimately keep next to their images, such as the Cisco IOU
keygen script, made every scan report errors and the sync job partial.
Extend the existing ignore filter with the extensions of scripts,
documentation and certificates. They are skipped by both the scan and
the file watcher, so they no longer trigger rescans either. Real
corrupted images are still reported.
The aiohttp URL builder only accepts strings, ints and floats as query
values and raises a TypeError on a Python bool, so deleting an image
answered a 500. Send true/false like the controller does.
GET /computes/{compute_id}/docker/images now returns the daemon image
id, size and creation date in addition to the tag, so a UI can show
and compare images instead of bare names. The fields are optional to
keep the response compatible with older computes.
Add DELETE /computes/{compute_id}/docker/images/{image} which removes
the image from the compute Docker daemon. The image name is passed as
a path so names with a slash or a tag separator work. A daemon that
refuses because a container uses the image answers a conflict; the
force query parameter removes it anyway.
The compute also drops the controller digest to local id translation
recorded when the image was synced, since it no longer matches
anything on the daemon.
getUrl joins the path directly after /v3/compute, so a path without
the leading slash produces /v3/computedocker/images/load and the
compute answers 404. Match the path shape used by forward() and the
image sync path.
Add POST /computes/{compute_id}/docker/images/load. It streams an
uploaded docker save tar (plain or gzip) to the compute, which loads
it into its Docker daemon. Clients can now install image packages
such as iol-xe-17-18-02.tar.gz without SSH access to the server.
The body is proxied without buffering (async iterable to chunked
transfer) and runs with no total timeout, like the existing image
sync path between computes.
Creating an ACE printed every route path of the application to stdout.
The debug loop was left behind when the route iteration moved to
_iter_route_paths() in 021991410.
Add GET /access/users/me/privileges. It returns the effective privileges
of the logged in user, flattened across all roles assigned to the user
(directly or through group membership). Superadmins get the full
privilege list.
This gives clients a way to render their UI based on permissions. The
server stays the authority for permission checks: a privilege in the
list can still be denied on a specific resource path.
duplicate() kept the original author in created_by. The project list
filters on created_by, so a non-admin user could not see the copy they
just made. Pass the caller through both duplication paths, the same way
import already does since #3017.
Projects created before GNS3 3.1 have no created_by field. The project
list filtered them out for every non-superadmin user, so all old
projects disappeared from the list after an upgrade.
Fall back to the pre-3.1 ACE-only visibility rule when created_by is
not set. Projects owned by another user stay hidden.
pytest 9.2 reads the old value of a monkeypatched attribute from the
instance __dict__. Starlette State keeps attributes in an internal
dict, so the old value looks unset and the teardown deletes the
attribute for the rest of the test session. The first test that
touches app.state.exiting passes but removes it, and every later
monkeypatch.setattr on it fails with AttributeError.
Use plain assignments instead.
By default uvicorn waits for open connections forever. A long-lived
streaming response keeps the server waiting after Ctrl+C. One common
case is the MCP Streamable HTTP transport: the client opens a
standalone GET request and the server answers with an SSE stream
(not the old SSE transport). A second Ctrl+C sets force_exit. Uvicorn
then skips the lifespan shutdown, so project and controller cleanup
does not run.
Add timeout_graceful_shutdown. After the timeout uvicorn cancels the
remaining tasks and runs the lifespan shutdown. Also make the four
notification WebSocket handlers stop when app.state.exiting is set.
The HTTP stream handlers already do this.
A field log showed one request stayed silent for 604s (the OpenAI SDK
default timeout), then the SDK retried on a fresh connection and got
200 OK in under a second. The abort signal did not cancel the in-flight
HTTP call, so the turn could look hung for ten minutes.
- set timeout (default 120s, request_timeout in the LLM config wins),
max_retries=1 and stream_usage=True for the providers that accept them
(openai, deepseek, anthropic, xai). ollama and google-genai reject
these kwargs, so they keep the old behavior.
- log the elapsed time of each LLM call. A slow first token or an SDK
retry now shows up in the log line.
- add request_timeout to the LLM config schemas and docs (JSON column,
no migration).
- tests: cover the defaults, the override, invalid values, provider
filtering and the title model.
The server link schema has show_filters_icon (default true) and the PUT
passes it through, but the MCP tool docs did not mention it. The old
copilot packet filter tool always set it to false to hide the filter icon
during fault injection; after the move to link_update the icon stayed
visible because the model had no way to know the field exists.
- link_update: add show_filters_icon to the supported kwargs
- link_list: add show_filters_icon to the available response fields
The graph passed remaining_steps=20 (50 for injection) in the run inputs.
RemainingSteps is a LangGraph managed value derived from recursion_limit,
so the input was ignored and the real budget was 25 (100 for injection).
No one could see those numbers, and a turn that hit the cap ended on a
bare tool result with no explanation.
- Remove the dead remaining_steps input.
- Add resolve_recursion_limit(): max_steps from the LLM model config wins;
otherwise 25 teaching_assistant, 50 lab_automation_assistant, 100
troubleshooting_injection.
- When fewer than 4 steps remain after a tool run, route to the new
budget_wrapup node. The next llm_call binds no tools and gets a wrap-up
instruction, so the turn ends with a short summary of what was done and
what remains.
- should_continue skips the title call after a wrap-up (no step left).
- budget_exhausted is reset at the start of every user turn.
- Add max_steps to the LLM config schemas. The config is a JSON column,
so no migration is needed.
- Tests: 16 new tests cover the resolver, the routing, the tool-free
wrap-up call and the removed input. Docs updated.
The bridge made 49 project-scoped MCP tools available, but the mode
tables still mirrored the old copilot capability set. Expand them:
- lab_automation_assistant: node_console_exec (raw console for nodes
without a device_type tag), node_list, snapshot_list/create/delete/
restore
- teaching_assistant: node_list/node_get, link_list, link_marker
(read-only diagnostics plus traffic-insight markers)
- troubleshooting_injection: link_capture_start/stop/download and
marker_definition (inject -> capture -> verify loop)
Also fix the tools_by_name registry: it was built from the lab and
injection tables only, which worked while teaching was a strict subset
of lab. Teaching-only tools would have raised KeyError in tool_node
when the LLM called them; the registry now unions all three mode
tables. A new test pins each mode's exact toolset, checks that no
administration tool leaks in, and asserts every referenced name is
resolvable.
The copilot and the MCP service maintained two parallel wrapper layers
(LangChain BaseTool classes vs @mcp.tool() functions) over the same
handlers in gns3_client/api_handlers.py. Descriptions and parameters
drifted apart (compute_id and link_type were missing on the copilot
side). This makes the MCP registration the single source of truth and
removes the duplicated wrappers.
tools_v2/mcp_bridge.py adapts the MCP server's tool registration into
LangChain tools: names, descriptions and argument schemas come from the
MCP tools. The pool is filtered to project-scoped tools
(MCP_ALLOWED_TOOLS) because the copilot runs inside one open project —
project management, compute/appliance/image administration and server
info tools stay out. The bridge carries the copilot's request JWT into
the MCP contextvars (username/token_version decoded from the token so
ticket-minting tools work) and runs the async MCP functions from the
copilot's sync tool node on either thread flavor.
Mode toolsets (teaching / lab_automation / troubleshooting_injection)
are now composed from adapted MCP tools plus a few copilot-native ones
that add value the MCP versions lack: link_create (port-name
resolution), wait_seconds, packet_analysis(+skills), and the topology
reader. Eight orphaned LangChain wrappers are deleted.
The device command tools had the dependency inverted (the MCP handlers
instantiated the copilot LangChain classes and round-tripped through
JSON strings). The Nornir/Netmiko classes moved verbatim (git mv) into
the shared layer as plain runners — DeviceConfigCommandRunner,
DeviceShowCommandRunner, VpcsCommandRunner — with _run(json_string)
becoming run(params, gns3_ctx). The optional Jinja2 template rendering
moved from the MCP handlers into the runners (and into
utils/device_configs.render_device_template), so both entry paths get
it; the MCP handlers are now pure delegation. This also drops the
json.dumps/json.loads detour the MCP path used to take.
test_tool_params consistency gate extended: a handler that forwards the
whole params dict to a callee is a wildcard (same rationale as the
params.items() case) — the per-key check cannot see through delegation.
Rename five copilot tools to the MCP <object>_<verb> naming so both
agent surfaces expose the same names:
- create_gns3_node -> node_create
- create_gns3_link -> link_create
- create_gns3_text_drawing -> drawing_create_text
- create_gns3_rectangle_drawing -> drawing_create_rectangle
- create_gns3_ellipse_drawing -> drawing_create_ellipse
Also close two capability gaps found during the comparison:
- node_create: accept optional per-node compute_id (passed through to
the shared handler; before, every node landed on local)
- link_create: accept optional per-link link_type (ethernet default,
serial supported; before, serial links were impossible)
All tools keep calling the same shared handlers in
gns3_client/api_handlers.py, so the rename has no effect on execution
semantics. Old names appeared nowhere else (prompts, tests, mode tool
lists are clean); docs updated accordingly.
Field finding from the smoke test: login flushes typed-ahead input
before printing "Password: ", so a password sent in a separate call is
lost (Login incorrect loop). Sending username and password in the same
command string logs in reliably; the shell then persists across calls.
websocat's text mode (-t, and also the default) appends a newline after
every received WebSocket message and turns CR/LF inside a message into
spaces. The compute ships console output as one BINARY frame per
arbitrary read(1024) burst, so text mode diced the stream at every frame
boundary — prompts split into pieces, short "narrow-column" lines, lost
carriage-return redraws (field report: Password: broken in three).
Measured against a local WS server sending known frames:
-t: A\nB\nC\na b\na b\n ("a\rb" -> "a b")
-b: ABCa\rba\r\nb (byte-exact, both directions)
Binary command frames stay console data on the compute side (only a line
that is itself a JSON {cols,rows} object would be taken as a resize
frame). The regression test fails on -t and passes on -b.
Replaces node_console (which handed a URL + ticket to the client for a
bash websocat chain) with a tool that runs the whole session server-side:
spawn `timeout N websocat -t --no-close <console-ws-url>`, feed the
command lines through stdin (paced, CRLF), read stdout until the timeout
kills websocat, return the collected output as plain text.
- plain text replies: "output: ..." or bare error text — no JSON
envelope, no status field (the server cannot judge command success)
- raw output: UTF-8 decode with replacement only, no telnet/ANSI
stripping; control keys are real characters (description teaches
\u0003, no \x03 translation)
- output funnel: grep=<python regex> filter; <=50 lines return as-is;
beyond that a 10-line sample plus hints — regrep=true re-filters the
node's cached capture (no re-run), full=true returns the last 400
lines / 64 KiB, and the full capture is written to the node's files
(console-capture.txt) with a ticketed curl download URL
- get_node_console_info_handler removed with its hashlib/TTL imports;
the node-bound console ticket redemption path stays (comment updated,
no current producer)
The skills repository carries per-protocol fault-injection knowledge
(symptoms, injection commands, verification steps) that only the
copilot could read through InjectionSkillsTool. Add the MCP counterpart
with the same multi-call protocol:
list (context REQUIRED) -> category index -> single fault
The list step is topology-filtered: context matches a skill's category,
key and declared protocols (case-insensitive), so aliases such as ebgp,
ospfv3 and stp select the right category instead of relying only on key
substrings. Fault bodies stay gated behind detail/issue requests.
Fault category keys are derived from repository filenames
(injection_<file>_issues). Correct the stale injection_ospf examples in
the registry docstrings, the loader comment and the copilot docs, which
pointed at keys that never existed.
Docs: 84 tools; new Fault Injection Skills section.
SkillsManager refused to start when the local skills directory was not a
git repository (cloning into a non-empty path fails), and reload_skills()
returned early when injection/ was missing, so device and feature skills
never loaded. reload_packet_analysis_protocols() had the opposite bug:
an empty result wiped the registry while still reporting success.
- initialize(): accept a non-empty directory without .git as a local-only
source (skip automatic updates, keep loading files); degrade to local
files when GitPython is unavailable.
- _update_if_needed(): re-create/clone when the local directory vanished
or moved, instead of reusing a stale handle forever.
- reload_skills(): injection, device and feature load independently; a
missing category keeps its previously loaded entries.
- reload_packet_analysis_protocols(): an empty/missing directory keeps
the existing registry and reports failure.
- reload_skills_repository(): drop the console guard cache so
config/console_guard.txt applies on reload.
MCP device_skill_get now exposes the category filter, documents the
single-file (topic-less) layout, and returns a sections list for
index/summary of skills that have no topics/issues. An uninitialized
skills repository fails with a clear error instead of an empty list.