-
Notifications
You must be signed in to change notification settings - Fork 276
ads: fix wrong accountability of sent resources #3803
Conversation
046e6a6
to
d758e9f
Compare
Codecov Report
@@ Coverage Diff @@
## main #3803 +/- ##
==========================================
- Coverage 67.76% 67.61% -0.15%
==========================================
Files 199 201 +2
Lines 11096 11337 +241
==========================================
+ Hits 7519 7666 +147
- Misses 3528 3620 +92
- Partials 49 51 +2
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
d758e9f
to
63ca157
Compare
I will add a couple unit tests for the newly added functions |
63ca157
to
a460516
Compare
This commit unblocks #3799 (needs rebase), which will fix #3775. We couldn't do a proper Istio's resource diff is not a superset (I took that from go-control-plane) and is in fact an |
a460516
to
a071c94
Compare
a071c94
to
1297704
Compare
This commit introduces 2 related changes: - Knowledge of what resources a certain proxy has requested/subscribed to, up to a current point in time and, - the use of this new data structure to avoid accounting for resources sent that might have not been requested by envoy. There's a potential bug in how we treated resources sent, where a non-wildcard vertical sending more resources than the requested/subscribed ones could make a later request be understood as an ACK. Related to openservicemesh#3775 Signed-off-by: Eduard Serra <[email protected]>
1297704
to
0cf7acf
Compare
PR openservicemesh#3803 introduced a mapset (backed by golang map) to slice conversion, which does not guarantee same output ordering for the same input, thus was making a unit test fail from time to time. This commit forces output of the slice to be alphabetically ordered, ensuring determinism of the output slice for the same input mapset. Signed-off-by: Eduard Serra <[email protected]>
PR openservicemesh#3803 introduced a mapset (backed by golang map) to slice conversion, which does not guarantee same output ordering for the same input, thus was making a unit test fail from time to time. This commit forces output of the slice to be alphabetically ordered, ensuring determinism of the output slice for the same input mapset. Signed-off-by: Eduard Serra <[email protected]>
PR openservicemesh#3803 introduced a mapset (backed by golang map) to slice conversion, which does not guarantee same output ordering for the same input, thus was making a unit test fail from time to time. This commit forces output of the slice to be alphabetically ordered, ensuring determinism of the output slice for the same input mapset. Signed-off-by: Eduard Serra <[email protected]>
This commit introduces 2 changes:
to, up to a current point in time and,
sent that might have not been requested by envoy yet, as well as using them
to generate stub requests from ctl-driven proxy updates.
There's a potential bug in how we treated resources sent, where a
non-wildcard vertical sending more resources than the requested/subscribed ones
could make a later request be understood as an ACK.
Consider the following case:
Related to #3775
Signed-off-by: Eduard Serra [email protected]
Please answer the following questions with yes/no.
No
No