Bazel Recipes @ Tink

From the bazel build //stockholm/...
How Bazel helps in streamlining the Tink development workflow

Transcript

1. Bazel Recipes @ Tink
2. //introduction meetup_presenter( name = “Kiril Videlov”, work = “@tink//:software_eng”, exports = [ “//bazel/presentation” ], twitter = “@krlvi” ) #Bazel contributor, commit hash: d96e1cdbe71d117f76e55c668bb4a0f976bf0bc0
3. //background • Tink is a fast growing company -> Colleagues & Code • Facilitate high rate of change • Quality for end users
4. Rate of change vs Quality As per convention: “Integrate small & often with as many tests as possible” But those build times...
5. Build & Test only targets that were affected by a change Most changes are small and isolated It would be great not to have to rebuild the world Binary compatibility is tricky What if we could...
6. Bazel (just water it regularly)
7. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame)
8. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Fully specified HD*(Hermetic & Deterministic) java_library( ) java_binary( ) go_library( ) container_image( ) inputs -> inputs -> inputs -> inputs -> inputs -> inputs -> inputs -> inputs -> inputs -> -> outputs -> outputs -> outputs -> outputs
9. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Dependencies graph Target’s inputs and outputs form a Directed Acyclic Graph Example:
10. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Reproducibility java_library( name = "lib", src = ["Foo.java", "Bar.java"] deps = ["//db:client"], ) • Construct a Merkle (hash) tree from inputs • Build in a sandbox* of only the defined inputs • Name artifact after the hash
11. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Cashing Artifact are cached locally. … But sharing is caring Remote caching speeds up builds
12. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Parallel execution • Targets can be built in parallel • Extra tricks - compile ABI first • Remote execution Example:
13. //Bazel semantics Isolation of Targets Dependency DAG Reproducibility Parallel Artifact execution caching Parsimony (endgame) Parsimony parsimony (n.) early 15c., from Latin parsimonia "sparingness, frugality, thrift," from pars-, past participle stem of parcere "to spare, save, refrain from, use moderately" (which is said to be unrelated to Latin parvus "small," parum "too little") + -monia, suffix signifying action, state, or condition. A.k.a.: Only build what is necessary Option A: Hit the cache Option B: Query the graph
14. //Option A: Hit that cache
15. //Option B: Query the graph Ingredients: • List of files created, updated or removed • Bazel Prep: • With the list of files, query for packages ◦ `bazel query “path/to/file”` • With the list of packages, query for targets of interest ◦ `bazel query ‘kind(docker_push_rule, allpaths(//…, packages))’` Serving: • Warm, with a pinch of CI
16. Q&A