[MRESOLVER-584] JDK transport HTTP/2 GOAWAY improvement #532
Conversation
|
I have a feeling that "swapping" will not even happen. What actually happened in this PR is that HttpClient got much shorter lifespan. Originally, HttpClient was kept and reused (for same session+repository) across whole session. Now, the client shares lifecycle with Transport instance, that is OTOH created by connector per invocation. |
cstamas
changed the title
|
This fix is wrong. Tested using reproducer https://issues.apache.org/jira/browse/MNG-8145 (reused same config and all). BUT, the reason why it works is due this PR "shortening" the lifespan of JDK HttpClient, and not due "client swap". When I set nginx config from original Still, the PR is correct in a way it keeps HttpClient per Transporter instance, and that is correct change. Will see in some other solution... |
|
Am removing "swap" bit, and will make this pr same as #533 |
cstamas
changed the title
Changes: * drop 21 closer, it abruptly uses "shutdown now" * in 11 make httpClient NOT stored per session, but per transport instance * make client "swappable", with rudimentary GOAWAY detection * simplify and refactor, put things in places like insecure mode config --- https://issues.apache.org/jira/browse/MRESOLVER-584
fire a thread for that that will eventually die off.
cstamas
force-pushed
the
MRESOLVER-584-jdk
branch
from
df02b97
to
ea72bf6
Compare
|
@scholzi100 am about to merge this one, as I changed JIRA goal also from "fix" to "improve", which this PR does. And later we can refine things even more. This improvement is in fact very welcome simplification, and the improvement part is basically shortening the lifespan of client. |
Changes:
https://issues.apache.org/jira/browse/MRESOLVER-584