Repository navigation
Weblab v0.13.1
Agent-driven failure tests, bulk console commands and native ARM64 container labs.
Native ARM64 source builds
Linux ARM64 hosts can now build and run the core, custom FRR routers, LL2S
switches and Alpine PCs natively using compose.arm64.yaml. FRR includes Weblab's
LLDP, read-only SNMPv2c, routed VLAN and IPv4 VRRP extensions. The builders select
native bases and pinned public sources; KVM and x86 emulation are not required.
Follow the ARM64 setup guide
to build the device images before starting Weblab. An
OSPF/RSTP practice lab
includes two routers, two switches and two PCs.
This is a source-build profile. The installer, GHCR release images and binary
release archives remain Linux amd64. Vendor IOL/QEMU guests and the optional
Agent, Internet and external SSH helpers are outside this initial ARM profile.
The UI offers native image selectors and rejects unsupported starts. ZIP restore
still requires exact image identity; use saved configuration text on fresh nodes
when moving between amd64 and ARM64.
Agent-driven failure scenarios
- New guarded tools let an Agent read cable state, apply or clear directional
frame loss, and request unplug/reconnect on supported endpoints. The separate
Allow link faults permission defaults off. Each change needs a fresh read
and approval, unless session YOLO explicitly preapproves the enabled action. - Topology, runtime and fault revisions prevent stale actions. Carrier and frame
loss remain independent: clearing a frame block does not reconnect a cable.
The Agent must verify resulting device behavior and report remaining faults. - A native MSTP exercise verified cable failure on SW2–SW4, SW4 rerouting through
SW5, and recovery after reconnect. All 12 directed PC connectivity checks
passed after failover and restoration. The transition probe lost 10 of 40
packets; it did not measure an exact convergence time.
See Agent permissions and the failure-test example
and link controls.
The integrated Agent is an optional self-hosted feature, unavailable in the public sandbox.
Console reliability and bulk commands
- Console reads, command helpers and named keys give Agents clearer prompt,
boot, pager and pending-input guidance. Unread output, stale cursors, expired
turns and human input locks remain guarded. A successful send is not proof
that a command finished or succeeded. - Layout → Bulk console commands…, also available in console Actions,
sends a reviewed command sequence to selected consoles. Agent/MCP bulk input
supports up to eight targets, with fresh reads, compatible platform/prompt
states and exact approval. It reports partial or uncertain delivery without
automatically replaying the batch. - Vendor guidance covers pagination and common console preparation. It does
not make every model reliable; inspect output and use a suitable model for
unattended work within your enabled permissions.
See bulk console commands
and the Agent console guide.
Reset and Junos saved state
- Workspace reset now clears Instructions by default. Check Preserve
Instructions to retain them. Reset guards include the document revision,
and other browser stations/pop-outs receive the updated state. - Stop node/Stop lab request graceful shutdown for the exact tested vJunos-switch
and vJunosEvolved 26.2R1.7 images. A failed request or 180-second timeout keeps
the VM running for inspection. Wait for stopped state before exporting or
restarting the core; stopping a container is not the interactive guest
shutdown workflow. - Real Junos ZIP restore, subsequent boot and forwarding were verified for
both tested profiles. Stop does not commit candidate configuration.
See reset and backup semantics
and Junos shutdown behavior.
Upgrade and validation
Use matching 0.13.1 core, Agent, EXOS and optional Internet helper versions
on amd64. Save device configurations, stop the lab and wait for shutdown before
recreating core. Recreate Agent with its existing dedicated volumes to obtain
the new tools while retaining history and sign-in state; reload the browser.
Link-fault permission remains off until explicitly enabled. The amd64 FRR and
LL2S releases remain 10.7.1-wl1 and 0.2.0-ovs-rstp1.
Validation includes backend, browser, MCP and packaged-container checks; native
Junos restore/shutdown and the Agent cable-failure exercise; and native Ampere
builds with OSPF, RSTP, LLDP/SNMP, PC forwarding, capture decoding, cable recovery
and ZIP restore. Release CI checks staged artifacts before promoting versioned
images. These checks do not certify every device image, protocol or model.
The v0.13.0 tag was not promoted: CI exposed a browser-test timing race when
measuring a newly opened console menu. v0.13.1 waits for the asynchronous layout
before checking viewport bounds. Runtime behavior is unchanged by that test fix.