How should I practice Feature Flag Evaluation Engine?

Start by reading the full statement, solve the visible examples, then compare the pattern against related Cresta onsite interview problems in FastPrep.

Is Feature Flag Evaluation Engine associated with Cresta?

FastPrep tags this as a Cresta onsite interview coding practice problem when the problem catalog has that company and stage context.

Problem · Design

Feature Flag Evaluation Engine

Learn this problem

● Hard

CrestaFULLTIMEONSITE INTERVIEW

Problem statement

Implement a feature flag evaluator over normalized rows derived from flag and user JSON data.

The inputs have these forms:

Each flags row is [flagKey, enabled, defaultValue], where the last two values are "true" or "false". Flag order is significant.
Each rules row is [flagKey, ruleId, value, rolloutPercentage]. An empty percentage means no rollout gate; otherwise it is an integer from 0 through 100.
Each conditions row is [flagKey, ruleId, attribute, operator, operand]. All conditions belonging to one rule are ANDed. A rule with no condition rows passes its condition group.
Each users row is [userId, attribute1, value1, attribute2, value2, ...].
Each queries row is [userId, flagKey]. The special key "*" requests all flags for that user.

Condition Operators

eq: the user value equals operand exactly.
in: the user value equals one token in the comma-separated operand list.
regex: the user value fully matches the valid anchored regular expression in operand.
~=: the lowercased ASCII operand is a substring of the lowercased ASCII user value.

An absent user attribute makes its condition false. Judged user values are always non-null strings.

Percentage Rollout

For a rollout rule, build flagKey:ruleId:userId and compute its unsigned 32-bit FNV-1a hash: start at 2166136261; for each ASCII byte, XOR the hash with the byte, multiply by 16777619, and reduce modulo 2^32. The rule passes its rollout gate exactly when hash mod 100 < rolloutPercentage.

Decision Rules

A disabled flag returns its default and ignores every rule.
For an enabled flag, a rule matches only when its condition group and optional rollout gate both pass.
If no rule matches, return the flag default.
If one or more rules match, return true when any matched rule has value true; otherwise return false.

Return one string per query. A single-flag query returns flagKey=true or flagKey=false. An all-flags query returns those pairs in input flag order, joined by |.

Function

evaluateFeatureFlags(flags: String[][], rules: String[][], conditions: String[][], users: String[][], queries: String[][]) → String[]

Examples

Example 1

flags = [["new_nav","true","false"],["beta_banner","false","true"]]rules = [["new_nav","country_rule","true",""],["new_nav","plan_rule","false",""],["new_nav","vip_rule","true","100"]]conditions = [["new_nav","country_rule","country","eq","US"],["new_nav","plan_rule","plan","in","free,trial"],["new_nav","vip_rule","email","regex","^vip.*@example[.]com$"]]users = [["u1","country","US","plan","free","email","regular@example.com"],["u2","country","CA","plan","trial","email","vip42@example.com"],["u3","country","CA","plan","paid","email","guest@example.com"]]queries = [["u1","new_nav"],["u2","*"],["u3","new_nav"]]return = ["new_nav=true","new_nav=true|beta_banner=true","new_nav=false"]

For u1, the country rule returns true while the plan rule returns false, so true-wins yields true. For u2, the plan rule and the 100% VIP rule both match, again yielding true; disabled beta_banner returns its default true. No rule matches u3, so new_nav returns its default false.

Example 2

flags = [["search","true","false"]]rules = [["search","platform_team","true",""]]conditions = [["search","platform_team","team","~=","PLATFORM"]]users = [["u1","team","Core Platform"],["u2","team","Mobile"]]queries = [["u1","*"],["u2","search"]]return = ["search=true","search=false"]

The ~= operator ignores ASCII letter case. Core Platform contains PLATFORM, while Mobile does not.

Example 3

flags = [["gradual","true","false"]]rules = [["gradual","r1","true","30"]]conditions = []users = [["u1"],["u2"]]queries = [["u1","gradual"],["u2","gradual"]]return = ["gradual=false","gradual=true"]

The FNV-1a buckets for gradual:r1:u1 and gradual:r1:u2 are 64 and 21. Only 21 is below the 30% threshold.

Constraints

1 <= flags.length <= 50
0 <= rules.length <= 1000
0 <= conditions.length <= 5000
1 <= users.length <= 10000
1 <= queries.length <= 500
Every flag key, rule ID, user ID, attribute name, user value, operand, and pattern is valid non-null ASCII text of length from 1 through 100.
Flag keys, rule IDs within a flag, user IDs, and attribute names within a user are unique. Flag keys contain neither *, |, nor =.
Each flag row, rule row, condition row, user row, and query row has exactly the shape described above, and every reference names an existing flag, rule, or user.
Every boolean field is exactly "true" or "false".
Each rollout percentage is empty or a base-10 integer from 0 through 100.
Every in operand is a comma-separated list of non-empty tokens, and its compared user value contains no comma.
Every regex operand begins with ^, ends with $, and is valid with the same full-match meaning in Python, Java, and C++14 ECMAScript regex; lookarounds and backreferences are excluded.