Skip to main content

Recording a diagnosis with Data Miner

Mod version shipped with FlammiCraft: 1.13.0

Deutsch · Reports and interpretation

Data Miner records diagnostics rather than repairing worlds. Enabling it causes automatic file output at the next process start.

Physical clients, including singleplayer, need the JVM argument -Ddata_miner.enabled=true. Dedicated servers enable it by default; false disables that process. Detailed method hooks require -Ddata_miner.detailedProfiling=true and add measurement overhead. Old config files are ignored; activation changes require restart.

Enabled launches automatically write nine supported registry dumps and startup reports into a new UTC/session directory. Basic tick monitoring and latency tracing start with a server. For a fresh performance section, permission-level-2 operators first use /data_miner perf stop and verify its export, then use /data_miner perf start, reproduce the problem and use perf stop to queue export. Starting an active performance recording reports that state and preserves its data. Events and latency have analogous start/stop commands. A queued response does not prove files finished writing; inspect logs and session_state.json. Caught export errors appear under export_worker.export_failures and prevent a completed session status.

Use dump to refresh that session's registries or mods analyze to regenerate analysis. Historical sessions remain under data_miner/info_client or info_server; JSONL timelines grow. Minute checkpoints limit some loss, but crashes/export failures can leave gaps. Dynamic biome/dimension registries are outside the supported dumps.

Compare similar workloads, durations and hook settings. A mod name in an observation is not proof of causation. Inspect source log text before sharing. The reference covers coverage, correlation and missing-data interpretation. No recording is started by this handbook work.

The existing read-only correlation command also interprets retained chunk requests, pipeline phases and admission envelopes. IDs join only within the same snapshot and recording. Repeated checkpoints stay separate and are never summed. Open envelopes do not distinguish dependency waiting, queue waiting, cancellation, failure or recording stop. Missing data remain unknown; the reference explains read limits and time mapping.