balancer/ringhash: fix uint32 weight sum overflow in normalizeWeights

[nvxbug]

newRing with two endpoints whose uint32 weights sum to 2^32 (e.g. an EDS locality weight of 2 times an endpoint weight of 2^30, which pass the per-locality and per-priority weight-sum checks):

ring.go: normalized endpoint weights is [{a +Inf ...} {b +Inf ...}]
ring.go: creating new ring of size 1024
(hangs)

normalizeWeights sums endpoint weights into a uint32, so a total above math.MaxUint32 wraps and can land on zero. The weight / weightSum divide then yields +Inf, newRing sets targetHashes to +Inf, and its for currentHashes < targetHashes loop never ends, growing the ring until the process is out of memory. Summing in uint64 cannot wrap and stays non-zero, since every endpoint weight is at least 1.

RELEASE NOTES:

• balancer/ringhash: Fix a bug where sum of endpoint exceeding math.MaxUint32 will get wrapped by storing the weight it uint64

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.55%. Comparing base (4793ad0) to head (d55cb28).
⚠️ Report is 9 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9357      +/-   ##
==========================================
+ Coverage   87.42%   87.55%   +0.12%     
==========================================
  Files         424      429       +5     
  Lines       30283    30622     +339     
==========================================
+ Hits        26476    26812     +336     
- Misses       3806     3809       +3     
  Partials        1        1              

Files with missing lines	Coverage Δ
balancer/ringhash/ring.go	100.00% <100.00%> (ø)

... and 44 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[eshitachandwani]

LGTM with 2 small comments about test comments. Adding @easwars as a second reviewer.

[eshitachandwani]

Nit : can we reword this a little to make it more readable , something along the lines of ... Tests the scenario where the sum of endpoint weights exceed math.MaxUint32 and would have been wrapped to zero using a unit32. Verifies that ring-build loop does not go on forever and produces a correct ring.
Or something more clear.

[eshitachandwani]

We need not write what the code used to do , rather we can write something like using uint32 can wrap it to a smaller value....