update envoy to 82b1f66

[snowp]

Signed-off-by: Snow Pettersen snowp@lyft.com

Description:
Risk Level: Medium
Testing: n/a
Docs Changes: n/a
Release Notes: n/a

[jpsim]

Have you verified the reflection required for envoyproxy/envoy#20169 is confirmed to work on iOS & Android? We use allow_multiple_dns_addresses for Happy Eyeballs which is one of the reloadable features we could check still works as expected.

Also, it looks like IdleTimeoutTests is crashing perhaps due to an uncaught exception.

[snowp]

I wouldn't expect there to be any platform specificity with that "reflection" since it's just reading the globally registered flags from absl instead of requiring a separate declaration (cc @alyssawilk). Is there an easy way to check that runtime value still work? Is there test coverage?

Re the ios test I think it's just a failing assertion, but the signal handler makes Envoy grab the ios exception and makes it look like a native one, looking into it

[snowp]

Yep removing Envoy signal handling I get

2022-03-16 07:48:24.886 xctest[29833:6895281] *** Assertion failure in -[XCTestExpectation fulfill], XCTestExpectation.m:77
Invalid connection: com.apple.coresymbolicationd
2022-03-16 07:48:25.136 xctest[29833:6895281] *** Terminating app due to uncaught exception 'NSInternalInconsistencyException', reason: 'API violation - multiple calls made to -[XCTestExpectation fulfill] for Str
eam idle timeout received by filter.'
*** First throw call stack:
(
        0   CoreFoundation                      0x000000010d169ba4 __exceptionPreprocess + 242
        1   libobjc.A.dylib                     0x000000010d01dbe7 objc_exception_throw + 48
        2   Foundation                          0x000000010db81aa2 _userInfoForFileAndLine + 0
        3   XCTestCore                          0x000000012592e97a -[XCTestExpectation fulfill] + 518
        4   test                                0x000000015143939a $s023test_swift_integration_A4_lib16IdleTimeoutTestsC0aeF0yyF0eF16ValidationFilterL_C7onError_11streamIntely5Envoy0nK0C_AI011FinalStreamM0CtF + 4
26
        5   test                                0x0000000151439628 $s023test_swift_integration_A4_lib16IdleTimeoutTestsC0aeF0yyF0eF16ValidationFilterL_C5Envoy08ResponseI0AagHP7onError_11streamIntelyAG0jM0C_AG011F
inalStreamO0CtFTW + 24
        6   test                                0x0000000151481305 $sSo15EnvoyHTTPFilterC0A0E6filterAbC6Filter_p_tcfcys6UInt64V_SSs5Int32VSo18envoy_stream_intelaSo0g7_final_h1_I0atcfU5_ + 549
        7   test                                0x0000000151482e92 $sSo15EnvoyHTTPFilterC0A0E6filterAbC6Filter_p_tcfcys6UInt64V_SSs5Int32VSo18envoy_stream_intelaSo0g7_final_h1_I0atcfU5_TA + 50
        8   test                                0x00000001514793a0 $ss6UInt64VSSs5Int32VSo18envoy_stream_intelaSo0c7_final_d1_E0aIegygyyy_ABSo8NSStringCAdfHIeyByyyyy_TR + 400
        9   test                                0x0000000152227d40 ios_http_filter_on_error + 416

We should probably disable this for ios tests

[snowp]

envoyproxy/envoy@d7edf3b this is the problematic change, it seems like changing what idle timeouts look like breaks our heuristic

[alyssawilk]

last time runtime screwed up we had a test fail so I think it'd detect that particular problem. it'd probably be good to have explicit rather than implicit tests for it though!