Skip to content

fix(web): share H2 initialization and join client closure #49

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
cppla merged 2 commits into from
Oct 3, 2026
Merged

fix(web): share H2 initialization and join client closure #49

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4c47c00
fix(web): share H2 initialization and join client closure
cppla Oct 3, 2026
d54d334
fix(web): retain H2 failures for queued callers
cppla Oct 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 19 additions & 4 deletions docs/WEB_COVER.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -380,15 +380,30 @@ IPv4 candidates and starts them with a short stagger; each candidate retains
its own UDP socket so a blackholed first address cannot consume the entire H3
budget before a working family is tried.

Source builds after v1.0.1 share the result of a cold H3 physical connection
Source builds after v1.0.1 share the result of a cold H2 or H3 physical connection
attempt with all callers already waiting on it. A failed handshake does not
make those callers start replacement handshakes one after another. A later
invocation may retry, subject to the existing `web-auto` cooldown policy.
H2 callers register before queuing for session selection, so they retain
that same completed failure even if it arrives before their selection turn;
new callers after publication can retry without an indefinite failure cache.
The shared attempt belongs to the client and retains its configured physical
dial timeout; canceling one caller stops only that caller's wait, not the
attempt needed by other callers. If all callers abandon it, the attempt can
continue until its existing timeout or client shutdown. Client Close cancels
and joins owned dialing and session-cleanup workers before returning. This
attempt needed by other callers. H2 keeps its TCP-connect timeout separate
from the existing TLS/H2 initialization budget; its bounded TLS retry after
HelloRetryRequest still uses the original remaining initialization budget.
If all callers abandon it, initialization can continue until its configured
timeout or client shutdown. A successful unclaimed H2 connection can remain
pooled, but does not send CONNECT or generate application authentication until
a live caller reserves its first stream. That caller alone owns bootstrap;
subsequent callers retain the existing short connection-bound credentials.
Client Close cancels and joins owned dialing and detached session-cleanup
workers before returning. Concurrent H2 Close callers wait for the same
completed closure and result, including late physical dial results. This
does not claim synchronous termination of every dependency goroutine; custom
TLS callbacks must return, and operations ignoring cancellation cannot be
forcibly interrupted. A failed TCP dial's non-nil connection is closed before
its failure is published; a nil successful result fails closed. This
does not change authentication, wire profiles, timeout defaults or sibling
stream ownership, and is not a general connection-speed or browser-equivalence
claim. The published v1.0.1 binary does not contain this change.
Expand Down
Loading
Loading