Fix crash related to dynamic updates to UDP proxy via CDS/EDS

[adams-shaun]

Ref: #26206

Commit Message:

We've been experiencing a crash similar to the issue reported above. Duplicating this scenario involved causing updates to the cluster HostSet via xDS.

When the CM posts the updates to workers, a new ThreadLocalCluster object is created and the previous one will be destructed soon thereafter. We expect a new ClusterInfo which uses this new TLC to be constructed and stored in the infos map; however, it seems that is not the case:

  // flat_hash_map::emplace()
  //
  // Inserts an element of the specified value by constructing it in-place
  // within the `flat_hash_map`, provided that no element with the given key
  // already exists.
  //
  // The element may be constructed even if there already is an element with the
  // key in the container, in which case the newly constructed element will be
  // destroyed immediately. Prefer `try_emplace()` unless your key is not
  // copyable or moveable.

Instead, I think this is a better option:

  // flat_hash_map::insert_or_assign()
  //
  // Inserts an element of the specified value into the `flat_hash_map` provided
  // that a value with the given key does not already exist, or replaces it with
  // the element value if a key for that value already exists, returning an
  // iterator pointing to the newly inserted element.  If rehashing occurs due
  // to the insertion, all existing iterators are invalidated. Overloads are
  // listed below.
  //

By using emplace with a pre-existing key, we actually do not create a new object and future calls that reference the stale ThreadLocalCluster (i.e. the call to info() down the onData() path) will cause segfault.

Additional Description:
Risk Level:
Testing:
Docs Changes:
Release Notes:
Platform Specific Features:
[Optional Runtime guard:]
[Optional Fixes #Issue]
[Optional Fixes commit #PR or SHA]
[Optional Deprecated:]
[Optional API Considerations:]

[repokitteh-read-only]

Hi @adams-shaun, welcome and thank you for your contribution.

We will try to review your Pull Request as quickly as possible.

In the meantime, please take a look at the contribution guidelines if you have not done so already.

🐱

Caused by: #33824 was opened by adams-shaun.

see: more, trace.

[adisuissa]

LGTM, thanks!
Please also add a release note.
Assigning @danzh2010 as codeownder.
/assign @danzh2010

[adisuissa]

style-nit: comments (here and below) end with '.'

[adisuissa]

LGTM, thanks!

[adisuissa]

/retest

[adisuissa]

LGTM, thanks for fixing this!
Can you please add a release note.
/assign-from @envoyproxy/senior-maintainers

[repokitteh-read-only]

@envoyproxy/senior-maintainers assignee is @lizan

🐱

Caused by: a #33824 (review) was submitted by @adisuissa.

see: more, trace.

[adisuissa]

LGTM, thanks!

[adisuissa]

This was approved by a senior-maintainer, merging.