Failure mode

[smarunich]

This pull request introduces several configuration improvements and dependency updates for the simulator, focusing on enhanced control over the KV cache and tokenizers, better event publishing, and improved validation and testing. The most significant changes are grouped below.

Configuration and KV Cache Enhancements

• Added new configuration options to Configuration struct: KVCacheSize, TokenBlockSize, TokenizersCacheDir, HashSeed, ZMQEndpoint, and EventBatchSize, with support for command-line flags, default values, and validation logic to ensure correct usage. Also, the code now reads HashSeed from the PYTHONHASHSEED environment variable if not set. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-56da73936108421d5a3311d2eb69b8649cd188148cab46940795e848d4e8dbeaR116-R129), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-56da73936108421d5a3311d2eb69b8649cd188148cab46940795e848d4e8dbeaR185-R188), [[3]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-56da73936108421d5a3311d2eb69b8649cd188148cab46940795e848d4e8dbeaR290-R301), [[4]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-56da73936108421d5a3311d2eb69b8649cd188148cab46940795e848d4e8dbeaR346-R353), [[5]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-56da73936108421d5a3311d2eb69b8649cd188148cab46940795e848d4e8dbeaR388-R394))
• Refactored blockCache to use the new configuration struct for initialization, including dynamic event batching, topic naming, and publisher setup. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-4f7e65f6f540700e74fb5fc1432132384518109ee31fa67aa7f548fcd39a1681L47-R64), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-4f7e65f6f540700e74fb5fc1432132384518109ee31fa67aa7f548fcd39a1681R221-R224))
• Improved block removal logic in the KV cache to correctly reference the block being removed. ([pkg/kv-cache/block_cache.goL131-R135](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-4f7e65f6f540700e74fb5fc1432132384518109ee31fa67aa7f548fcd39a1681L131-R135))

Dependency and Build Updates

• Upgraded github.com/llm-d/llm-d-kv-cache-manager to a newer version and updated github.com/daulet/tokenizers to v1.22.1 to match the imported version, including changes to the Dockerfile and Makefile for consistent tokenizer version handling. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-33ef32bf6c23acb95f5902d7097b7a1d5128ca061167ec0716715b0b9eeaa5f6L11-R11), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-33ef32bf6c23acb95f5902d7097b7a1d5128ca061167ec0716715b0b9eeaa5f6L30-R29), [[3]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-dd2c0eb6ea5cfc6c4bd4eac30934e2d5746747af48fef6da689e85b752f39557L26-R28), [[4]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-76ed074a9305c04054cdebb9e9aad2d818052b07091de1f20cad0bbac34ffb52R42-R51))
• Removed unused and obsolete dependencies from go.mod for a cleaner build. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-33ef32bf6c23acb95f5902d7097b7a1d5128ca061167ec0716715b0b9eeaa5f6L20), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-33ef32bf6c23acb95f5902d7097b7a1d5128ca061167ec0716715b0b9eeaa5f6L40), [[3]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-33ef32bf6c23acb95f5902d7097b7a1d5128ca061167ec0716715b0b9eeaa5f6L72))

Event Publishing Improvements

• Updated publisher implementation and tests to use msgpack/v5 and configure the encoder for array-encoded structs, improving event batch serialization and compatibility. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-d28058992f73559fb3a5fd1eae56eceb8548c0b1a31f83fee179cf1a6b4be482R20-R28), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-d28058992f73559fb3a5fd1eae56eceb8548c0b1a31f83fee179cf1a6b4be482L65-R70), [[3]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-d28058992f73559fb3a5fd1eae56eceb8548c0b1a31f83fee179cf1a6b4be482L76-R81), [[4]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-11e396f9556390402b5e8eb73136dd18add33faf873f48f08b9a5ce4ae9f2d62L28-R28), [[5]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-11e396f9556390402b5e8eb73136dd18add33faf873f48f08b9a5ce4ae9f2d62R47-R48))

CI/CD and Documentation

• Added a prerelease input to the Docker build GitHub Action and fixed the logic for handling pre-release builds. ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-76cc00033cb7774949c1149f83441a859634155ccf9d2bf6dec2794d4b0cc107R16-R19), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-76cc00033cb7774949c1149f83441a859634155ccf9d2bf6dec2794d4b0cc107L35-R39))
• Updated documentation (README) to clarify and comment out new KV cache-related configuration options. ([README.mdR119-R127](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-b335630551682c19a781afebcf4d07bf978fb1f8ac04c6bf87428ed5106870f5R119-R127))

Testing Improvements

• Added and updated tests for new configuration options, including validation for invalid values (negative cache size, invalid block size, negative event batch size). ([[1]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-b836ab7773c80e16c32843c9030f596048b5967467f691a35aae92f93125e71cR106-R112), [[2]](https://github.com/smarunich/llm-d-inference-sim/pull/1/files#diff-b836ab7773c80e16c32843c9030f596048b5967467f691a35aae92f93125e71cR286-R300))