Conversation
|
/assign @danzh1989 |
|
@danzh1989 cannot be assigned to this issue. |
This PR updates our QUICHE dependency to commit 88c8d5903d851744410ea9840201b6507feae981. Signed-off-by: David Schinazi <dschinazi@google.com>
a69206e to
51bd0da
Compare
|
/assign @danzh1989 |
|
@danzh1989 cannot be assigned to this issue. |
|
/assign @danzh1989 |
|
@danzh1989 cannot be assigned to this issue. |
moderation
left a comment
There was a problem hiding this comment.
Otherwise deps look OK
Signed-off-by: David Schinazi <dschinazi@google.com>
danzh2010
left a comment
There was a problem hiding this comment.
Thanks for helping with updating quiche deps!
Signed-off-by: David Schinazi <dschinazi@google.com>
Signed-off-by: David Schinazi <dschinazi@google.com>
Signed-off-by: David Schinazi <dschinazi@google.com>
bdd3571 to
83e82f9
Compare
Signed-off-by: David Schinazi <dschinazi@google.com>
|
/assign @htuch |
|
/lgtm deps |
|
@danzh2010 @DavidSchinazi @alyssawilk I'm able to crash Envoy after this update. Using Firefox nightly with a downstream h3 config I can now generate
Looks like only Firefox nightly triggers this. The other h3 clients are all working fine - https://dev.to/moderation/experiments-with-h3-clients-envoy-52gj |
|
Thanks for reporting this crash!
HCM filter is initialized after handshake complete. My guess is that this
update enables 0-RTT, which creates streams before handshake complete.
#15912 would fix the issue. But
meanwhile, we probably should disable 0-RTT support in the QUIC listener.
…On Fri, Apr 23, 2021 at 11:17 AM moderation ***@***.***> wrote:
@danzh2010 <https://github.com/danzh2010> @DavidSchinazi
<https://github.com/DavidSchinazi> @alyssawilk
<https://github.com/alyssawilk> I'm able to crash Envoy after this
update. Using Firefox nightly with a downstream h3 config I can now
generate terminated by signal SIGSEGV (Address boundary error) within a
couple of page reloads.
[2021-04-23 08:10:42.986][127974][error][envoy_bug]
[source/common/quic/envoy_quic_server_session.cc:50] envoy bug failure:
false. Details: Quic session 3175632122628430394 attempts to create stream
0 before HCM filter is initialized.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#16100 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADQ23C6EUT6EJLNNYCQBAXLTKGFQDANCNFSM43KUCNZQ>
.
|
|
Michael, can you also provide the crash stack?
…On Fri, Apr 23, 2021 at 11:38 AM Dan Zhang ***@***.***> wrote:
Thanks for reporting this crash!
HCM filter is initialized after handshake complete. My guess is that this
update enables 0-RTT, which creates streams before handshake complete.
#15912 would fix the issue. But
meanwhile, we probably should disable 0-RTT support in the QUIC listener.
On Fri, Apr 23, 2021 at 11:17 AM moderation ***@***.***>
wrote:
> @danzh2010 <https://github.com/danzh2010> @DavidSchinazi
> <https://github.com/DavidSchinazi> @alyssawilk
> <https://github.com/alyssawilk> I'm able to crash Envoy after this
> update. Using Firefox nightly with a downstream h3 config I can now
> generate terminated by signal SIGSEGV (Address boundary error) within a
> couple of page reloads.
>
> [2021-04-23 08:10:42.986][127974][error][envoy_bug]
> [source/common/quic/envoy_quic_server_session.cc:50] envoy bug failure:
> false. Details: Quic session 3175632122628430394 attempts to create stream
> 0 before HCM filter is initialized.
>
> —
> You are receiving this because you were mentioned.
> Reply to this email directly, view it on GitHub
> <#16100 (comment)>,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ADQ23C6EUT6EJLNNYCQBAXLTKGFQDANCNFSM43KUCNZQ>
> .
>
|
|
Can we just flag off 0-rtt for both upstream and downstream? I assume there's a flag for that and neither side appears to support it yet. |
|
The crash discussed above should be resolved by #16147 |
|
I suspect we already support 0-RTT upstream. The resumption ticket is
stored in client crypto config, so as long as our client reuse it upstream
client will get 0-RTT. But I need to test and verify that.
…On Mon, Apr 26, 2021 at 10:26 AM David Schinazi ***@***.***> wrote:
The crash discussed above should be resolved by #16147
<#16147>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#16100 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADQ23C6LQ7UFFKEKDAWCYO3TKVZ2TANCNFSM43KUCNZQ>
.
|
|
I can report that the crash still happens after #16147. To reliably reproduce I can pull up a listener in Firefox Nightly on Linux and load the page. On first load and subsequent immediate reloads the site is fine. But then if I step away for > 5 mins and reload, crash. |
This PR updates our QUICHE dependency to commit 88c8d5903d851744410ea9840201b6507feae981. Signed-off-by: David Schinazi <dschinazi@google.com> Signed-off-by: Gokul Nair <gnair@twitter.com>
This PR updates our QUICHE dependency to commit
88c8d5903d851744410ea9840201b6507feae981.