Release 0.36.1: a full connection cap delays the startup self-probe, it does not fail it #353
No reviewers
Labels
No labels
code-review
correctness
dos
performance
security
severity/high
severity/low
severity/medium
tech-debt
Kind/Breaking
Kind/Bug
Kind/Documentation
Kind/Enhancement
Kind/Feature
Kind/Security
Kind/Testing
Priority
Critical
Priority
High
Priority
Low
Priority
Medium
Reviewed
Confirmed
Reviewed
Duplicate
Reviewed
Invalid
Reviewed
Won't Fix
Status
Abandoned
Status
Blocked
Status
Need More Info
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
h-dv/code-index!353
Loading…
Reference in a new issue
No description provided.
Delete branch "hotfix-0.36.1"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
What broke (field report, zammad-dev, 12k-file Rails tree)
After the schema v80+v81 migrations (~3 min), the first daemon aborted 10 ms after its accept loop started:
Nothing served for ~25 min until a later daemon took over.
Cause: the daemon publishes its port before it migrates, so clients retrying during the migration queue on it. When the accept loop started, they took all 64 slots and the rest were closed over the cap — the self-probe among them.
probe_reachablemade ONE attempt, read EOF, and answered "unreachable" without using its 5 s budget. Pre-existing (one cap warning on 2026-10-10 under 0.35); v81 doubled the window.Fix
probe_reachableretries one-shothealthround trips until its deadline. A refused attempt proves the cap is full, not that the loop is wedged. The takeover probe of a live holder gets the same reading. Against a live holder that refuses connections, it now takes its full 2 s budget instead of answering at once.Evidence
server::preauth_tests::a_full_connection_cap_delays_the_self_probe_but_does_not_fail_it: 64 live peers fill the real accept loop's cap and leave after 500 ms. The probe must answer true, and only after they left.reachable. 30/30 green in parallel.-D warnings --document-private-items), nextest ci 5029/5029, ci-timing 34/34, daemon leg 919/919, fast_gates rc=0, guest_gates rc=0.version_bump.sh), no schema change.🤖 Generated with Claude Code
https://claude.ai/code/session_01YDDBdWU4yThi8Zi51E6FCj
Added: package download URLs (
876b78b5)Folded into 0.36.1 by operator decision. ruby 0.10.0's package release (run 1094) published its assets, but the package channel listed them at a URL that returns 404.
Measured on the host:
…/releases/download/pkg/de.h-dv.ruby/v0.10.0/…returns 404.…/releases/download/pkg%2Fde.h-dv.ruby%2Fv0.10.0/…returns 200, and is the API's ownbrowser_download_url.Three causes, three fixes:
distribution/registry.v1.jsonused it, andregistry_schema_gaterequired it.code_index_core::registry::own_release_segmentgives one encoded spelling, used by every producer and gate (catalog row, schema gate,move_catalog_row,release_registry.sh,version_bump.sh, publish DRY_RUN).pkg/row is now refused.build_channelreported a same-digest version asmerge_kept, so the old URL from the core catalog beat the host's answer.merge_released: same digest takes the incoming URLs and is reportedmerge_relocated.package_channel_check.sh --downloadablenow fetches the released version'surlandsignature_urlfrom the PUBLISHED channel, anonymously, and requires 2xx.package-release.ymlruns it.Verification (mine, beyond the lane's 12 mutations):
a_released_version_takes_the_hosts_urls_over_a_same_digest_row.--downloadablealways 200 → RED,the_channel_check_refuses_a_listed_url_that_does_not_download.urlandsignature_url).crates/core) does not check URL shape, so v0.36.0 clients accept the encoded URL. A v0.36.0plugin addfrom the encoded URL installs, as observed in the field.After the 0.36.1 release: re-run package-release for
pkg/de.h-dv.ruby/v0.10.0. With (b) it rewrites the channel's URL, and with (c) it proves the new one downloads.🤖 Generated with Claude Code
Shipped. Head
876b78b5is on master's first-parent line, because master was fast-forwarded to it rather than merged through this PR, and it is tagged v0.36.1. Closing as delivered.Pull request closed