Online edition for BLOGE
0.9.9-RC1· prepared 2026-11-02 · 中文
Appendix H — Durable and Scale Runtime Reference
Chapters 21 and 22 keep mechanisms and decisions in the reading path. This appendix keeps deployment checks and measurement parameters.
Lease and recovery configuration
Record owner identity, claim timeout, scan interval, batch size, maximum recovery attempts, and logical/wall-clock source. A shorter timeout is not automatically safer: long pauses and store jitter increase false takeover, so timeout must be paired with fencing epochs.
Routing and migration
ShardResolver chooses a location for a new business key; existing keys read RoutingStore. Migration order is: stop writes, copy complete state, validate, atomically update the binding, resume traffic. A new hash after scaling is not migration.
Remote dispatch parameters
| Parameter | Record with it |
|---|---|
| topic / queue | worker capability and version |
| heartbeat | false-positive window and redelivery policy |
| batch size | global in-flight ceiling |
| retry | effect idempotency and duplicate delivery |
| result timeout | cancellation, late result, and DLQ handling |
Store migration check
Use the Flyway scripts packaged with the selected bloge-durable-mybatis artifact as authority; an old chapter's V25 name is not a permanent contract. Per environment, record schema history, table/index presence, application artifact checksum, migration time, and rollback decision.
Benchmark record template
commit / artifact:
JDK / CPU quota / heap:
topology / operator behavior:
arrival model / concurrency:
warmup / measurement / forks:
external stubs and store:
raw result path:
single changed axis:
stop rule:
claim and non-claim:
benchmarks/run.sh and compare.py provide run and comparison entry points. Read thresholds from the current script or team baseline; copying a number from the book does not make it current.