runtime: simplifying absl-flags based implementation #20169

alyssawilk · 2022-03-01T19:36:37Z

moving from a statically constructed envoy-flag-name -> absl flag map to using reflection to construct an envoy-flag-name -> commandline flag map, to avoid having to add 2 lines per runtime guard.

Risk Level: Medium
Testing: existing tests cover
Docs Changes: n/a
Release Notes: n/a

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

repokitteh-read-only · 2022-03-01T19:36:40Z

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #20169 was opened by alyssawilk.

see: more, trace.

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

RyanTheOptimist

This looks awesome! It'd be great if the PR description could have a bit more description of what's changing. But this is great!

source/common/runtime/runtime_features.cc

RyanTheOptimist · 2022-03-08T14:57:01Z

source/common/runtime/runtime_features.cc

@@ -87,7 +133,8 @@ bool runtimeFeatureEnabled(absl::string_view feature) {
    IS_ENVOY_BUG(absl::StrCat("Unable to find runtime feature ", feature));
    return false;
  }
-  return absl::GetFlag(*flag);
+  // We validate in map creation that the flag is a boolean.
+  return flag->TryGet<bool>().value();


OOC: What is the difference between TryGet().value() and GetFlag()?

GetFlag takes the absl flag and returns a commandlineflag. Here we don't have the absl flag we only have the string name, so tryget allows us to get a coommandlineflag from that.

AH! I see. That makes sense. I think I got confused because link 138 didn't change so I assumed that the type of flag didn't change. I think this is an example of how auto can actually make code a bit less readable, but it's a tradeoff, for sure. For bonus points consider changing auto to absl::CommandLineFlag on line 138. But totally at your discretion.

RyanTheOptimist · 2022-03-08T14:58:37Z

source/common/runtime/runtime_features.cc

 void maybeSetRuntimeGuard(absl::string_view name, bool value) {
  auto* flag = RuntimeFeaturesDefaults::get().getFlag(name);
  if (flag == nullptr) {
    IS_ENVOY_BUG(absl::StrCat("Unable to find runtime feature ", name));
    return;
  }
-  absl::SetFlag(flag, value);
+  std::string err;
+  flag->ParseFrom(value ? "true" : "false", &err);


Interesting. The old code looked simpler, but I suspect it was deficient in some way. Can you elaborate?

the code is cleaner using absl::Flags, but while you can get from string to CommandLineFlag there's no public registry from string to absl::Flags - the reflection only goes from string to CommandLineFlag, so I had to switch from the "I know my type" absl::Flag to "am I a boolean or not?" CommandLineFlag variant to get rid of the array.

Makes sense. Same point about auto confusing me. Same bonus points for changing auto to absl::CommandLineFlag :)

test/test_common/test_runtime.h

test/test_listener.cc

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk · 2022-03-09T14:51:57Z

cc @envoyproxy/envoy-mobile-maintainers I think we've solved the reflection problems but ccing just in case.

RyanTheOptimist

Awesome! Thanks for the explanation. Two suggests to eliminate the use of auto which led to me misunderstanding the code. But you can take or leave those suggestions.

test/test_listener.cc

RyanTheOptimist · 2022-03-09T19:51:16Z

source/common/runtime/runtime_features.cc

@@ -87,7 +133,8 @@ bool runtimeFeatureEnabled(absl::string_view feature) {
    IS_ENVOY_BUG(absl::StrCat("Unable to find runtime feature ", feature));
    return false;
  }
-  return absl::GetFlag(*flag);
+  // We validate in map creation that the flag is a boolean.
+  return flag->TryGet<bool>().value();


AH! I see. That makes sense. I think I got confused because link 138 didn't change so I assumed that the type of flag didn't change. I think this is an example of how auto can actually make code a bit less readable, but it's a tradeoff, for sure. For bonus points consider changing auto to absl::CommandLineFlag on line 138. But totally at your discretion.

RyanTheOptimist · 2022-03-09T19:51:53Z

source/common/runtime/runtime_features.cc

 void maybeSetRuntimeGuard(absl::string_view name, bool value) {
  auto* flag = RuntimeFeaturesDefaults::get().getFlag(name);
  if (flag == nullptr) {
    IS_ENVOY_BUG(absl::StrCat("Unable to find runtime feature ", name));
    return;
  }
-  absl::SetFlag(flag, value);
+  std::string err;
+  flag->ParseFrom(value ? "true" : "false", &err);


Makes sense. Same point about auto confusing me. Same bonus points for changing auto to absl::CommandLineFlag :)

alyssawilk · 2022-03-09T19:56:37Z

-> Snow for non-googler pass

snowp

Some questions but otherwise this seems good to me

snowp · 2022-03-09T23:46:28Z

source/common/runtime/runtime_features.cc

-  }
-}
-
-void RuntimeFeatures::restoreDefaults() const {


Is this stuff no longer necessary?

yeah, it's replaced by the absl flagsaver infrastructure

snowp · 2022-03-09T23:47:53Z

source/common/runtime/runtime_features.cc

+// by runtime feature guards. If you add a guard of the form
+// RUNTIME_GUARD(envoy_reloadable_features_my_feature_name)
+// here you can guard code checking against "envoy.reloadable_features.my_feature_name".
+// Please note the swap of envoy_reloadable_features_ to envoy.reloadable_features.!


What happens if people mess this up? Will it error out? If not, can we do something to avoid this mismatch?

it'll work just fine actually, it just won't have standard envoy guard naming.

snowp · 2022-03-09T23:48:42Z

source/common/runtime/runtime_features.cc

+    if ((!absl::StartsWith(name, "envoy_reloadable_features_") &&
+         !absl::StartsWith(name, "envoy_restart_features_")) ||


Just so I understand since I'm not that familiar with this code: why this filtering? How are flags accessed if not via this?

this largely helps us avoid pulling in erroneous flags in google 3, where we may end up with non-envoy command line flags.

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

snowp

Thanks!

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

moving from a statically constructed envoy-flag-name -> absl flag map to using reflection to construct an envoy-flag-name -> commandline flag map, to avoid having to add 2 lines per runtime guard. Risk Level: Medium Testing: existing tests cover Docs Changes: n/a Release Notes: n/a Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

runtime: simplifying

df1754e

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk added 4 commits March 2, 2022 11:19

Merge branch 'main' into runtime_cleanup

5b56ea7

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

cleanup

c3deda3

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

rework

bf65934

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Merge branch 'main' into runtime_cleanup

44d38d6

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk force-pushed the runtime_cleanup branch from 748805e to 44d38d6 Compare March 7, 2022 19:10

alyssawilk marked this pull request as ready for review March 8, 2022 13:26

alyssawilk assigned RyanTheOptimist Mar 8, 2022

RyanTheOptimist reviewed Mar 8, 2022

View reviewed changes

alyssawilk added 2 commits March 9, 2022 09:39

Merge branch 'main' into runtime_cleanup

1ed7bb1

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

comments

88b6212

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk changed the title ~~runtime: simplifying~~ runtime: simplifying flags based implementation Mar 9, 2022

alyssawilk changed the title ~~runtime: simplifying flags based implementation~~ runtime: simplifying absl-flags based implementation Mar 9, 2022

RyanTheOptimist previously approved these changes Mar 9, 2022

View reviewed changes

alyssawilk assigned snowp Mar 9, 2022

snowp reviewed Mar 9, 2022

View reviewed changes

alyssawilk added 2 commits March 10, 2022 11:25

Merge branch 'main' into runtime_cleanup

a15fcae

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

less auto

c17ab29

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk dismissed RyanTheOptimist’s stale review via c17ab29 March 10, 2022 16:25

fix bad merge

b1ba0ad

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

snowp approved these changes Mar 11, 2022

View reviewed changes

Merge branch 'main' into runtime_cleanup

d3feb84

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk enabled auto-merge (squash) March 14, 2022 15:44

Merge branch 'main' into runtime_cleanup

d34f110

Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

alyssawilk merged commit c4fa707 into envoyproxy:main Mar 14, 2022

alyssawilk deleted the runtime_cleanup branch March 20, 2023 14:37

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

runtime: simplifying absl-flags based implementation #20169

runtime: simplifying absl-flags based implementation #20169

alyssawilk commented Mar 1, 2022 •

edited

Loading

repokitteh-read-only bot commented Mar 1, 2022

RyanTheOptimist left a comment

RyanTheOptimist Mar 8, 2022

alyssawilk Mar 9, 2022

RyanTheOptimist Mar 9, 2022

RyanTheOptimist Mar 8, 2022

alyssawilk Mar 9, 2022

RyanTheOptimist Mar 9, 2022

alyssawilk commented Mar 9, 2022

RyanTheOptimist left a comment

RyanTheOptimist Mar 9, 2022

RyanTheOptimist Mar 9, 2022

alyssawilk commented Mar 9, 2022

snowp left a comment

snowp Mar 9, 2022

alyssawilk Mar 10, 2022

snowp Mar 9, 2022

alyssawilk Mar 10, 2022

snowp Mar 9, 2022

alyssawilk Mar 10, 2022

snowp left a comment

		if ((!absl::StartsWith(name, "envoy_reloadable_features_") &&
		!absl::StartsWith(name, "envoy_restart_features_")) \|\|

runtime: simplifying absl-flags based implementation #20169

runtime: simplifying absl-flags based implementation #20169

Conversation

alyssawilk commented Mar 1, 2022 • edited Loading

repokitteh-read-only bot commented Mar 1, 2022

RyanTheOptimist left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alyssawilk commented Mar 9, 2022

RyanTheOptimist left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alyssawilk commented Mar 9, 2022

snowp left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

snowp left a comment

Choose a reason for hiding this comment

alyssawilk commented Mar 1, 2022 •

edited

Loading