Repository navigation
Lab/routing backtrack - #132
Merged
Merged
Conversation
…e no-systemd test is hermetic test_systemctl_without_systemd_is_not_systemd failed on any machine that really runs systemd, because detect_init looked at the real /run/systemd/system. The directory can now be overridden; the test points it at a path that does not exist. Version 0.4.89. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The first-run card, the standing bar and the topology legend all said the same thing about a node with no network, at length. The bar and the card are now one sentence and two instructions, and the legend waits for a link to explain. Version 0.4.90. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…dled exception on accept Measured on 40 to 200 in-process nodes: the first handshake to a node nobody links to is lost to a forwarding cycle the replay table breaks, and a shorter retry does not help. Recorded as open with the reproduction and the options. Version 0.4.91. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…r way A relay forwarded to the XOR-nearest neighbour except the link a packet came from, and the replay table dropped the second visit to a node: a cycle ate the only copy. A relay now remembers, per packet and bounded, the identities it arrived from and left to, and sends a duplicate that arrives over a new identity to a neighbour not yet tried. Replays over a known identity are still dropped, a packet leaves by at most the fanout, and the book is bounded. Measured on 120 nodes over eight processes: 285 of 300 first contacts arrive within 12 s instead of 256, and the sender retries 15 times instead of 45, for about 7 % more packets relayed. Adds tests/tools/mesh_sim.py, which spreads the nodes over processes and reports the event loop's lateness, after single-process figures at 100 and 200 nodes were found to measure the simulator. BUGSVULNS 100 is rewritten and the withdrawn figures say so. Version 0.4.92. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… of the accept callback The CHALLENGE send at the end of _on_new_transport was unguarded, unlike the capabilities announcement before it, so a far end that let go between the accept and our first words surfaced as an unhandled exception in the transport's callback. It now ends the link like any failed send. BUGSVULNS 101. Version 0.4.93. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… same node _authenticated_peers ran for every routed packet and scored every link, though a score only settles a choice between two links to one identity. A second link now triggers the scoring. On 120 nodes over eight processes that function went from 32 s to 2.4 s in one worker's profile and the workers' CPU at rest fell by about 16 percent over three alternating runs. mesh_sim.py gains --profile and reports the CPU the workers used. Version 0.4.94. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ould forward Measured on 100 nodes over eight processes: a fifth of first contacts are lost at 20 percent black-hole relays. A retry by another first hop and a quicker retry were tried and gave no gain in delivery, so they are recorded and not kept. mesh_sim.py learns --blackhole. Version 0.4.95. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…, not an interpreter message node was read once at the top of the fleet routes and never checked to be text, so an object or a list reached a dictionary lookup and came back as a 503 saying 'unhashable type'. It is now a 400, and a type error that gets past the checks is logged and answered without the interpreter's words. Adds tests/tools/api_fuzz.py, which signs in to a throwaway node and sends every route hostile requests; all 60 routes refuse a request without a session. BUGSVULNS 103. Version 0.4.96. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The console wraps its listening socket for TLS, so the handshake ran inside accept() and a client that connected and said nothing held the only accepting thread, with no timeout and no password needed. The three HTTP servers (console, chat, MCP) now share HardenedServer: the handshake is the connection's own, a silent connection is closed at 30 s, at most 256 are served at once, and a client that hangs up is not written to stderr (500 resets used to leave 723 kB of traceback in the log). Version 0.4.97. BUGSVULNS 104. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The total bound alone turned one client's unfinished requests into the console refusing everybody: 256 requests that never finish, a byte now and then, filled every slot. One address is now held to 32, measured against a real node (300 attempts left 32 open, and another machine was served). BUGSVULNS 104 says the first fix was not enough. Version 0.4.98. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…iptors follow the bind host A lan address was confirmed only by an inbound connection, so a node nobody had connected to refused its first short invitation and sent the operator to a three-block manual exchange. With no confirmed address and no relay it now offers an address on its own network, marked untried, and the page says so; a public address nothing has reached is still never offered. Found on the way: ip_reachability listed every address the machine holds whatever the listener was bound to, while expand_listen_uri honours a concrete bind host. A loopback listener announced LAN addresses it did not listen on. Descriptors now follow the bind host. BUGSVULNS 105 and 106. Version 0.4.99. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…e list Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… anyone without the key Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…orter identity text; ui_walk tool Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…nk past a soft limit Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
…reads it Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.