Skip to content

HashBurst Node Installer v2.1.4

Choose a tag to compare

View all tags
@hashburst hashburst released this 12 Aug 00:19
· 137 commits to main since this release
v2.1.4
8497c6c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

HashBurst Node Installer v2.1.4

v2.1.4 advances the native HashBurst replication controller from the v2.1.3 durable-replication foundation to a safe object-release lifecycle.

Safety defaults

  • Controller bind remains 127.0.0.1:8095 by default.
  • Controller mode remains observe by default.
  • Physical UNPIN requires controller HB_REPL_UNPIN_ENABLED=1.
  • The node agent independently requires HB_REPL_ALLOW_UNPIN=1.
  • Every destructive UNPIN also requires just-in-time authorization for the exact node, job, lease, CID and object generation immediately before local pin/rm.
  • Final logical release waits HB_REPL_DELETE_GRACE_SEC=900 seconds before an UNPIN can be scheduled.
  • The installer ships controller/agent units but does not automatically enable them.

Lifecycle additions

  • Durable logical-reference tombstones preserve idempotency after release.
  • Final release increments the object generation and marks existing replica desired state false.
  • Pending PIN/VERIFY work from older generations becomes stale.
  • Re-registration is rejected while an UNPIN has crossed the authorization boundary and its physical outcome is unknown.
  • Startup recovery converts an authorized-but-unconfirmed destructive operation into a verification path instead of assuming either success or failure.
  • Extra replicas can be trimmed only when the N total target and M committable minimum remain satisfied.

Packaging fix

hb_ipfs.py is installed into /opt/hashburst/replication/hb_ipfs.py as well as the HB-Files tree. This makes direct systemd execution of the replica agent self-contained and fixes the missing-module failure discovered during the v2.1.3 production rollout.

Failure domains

v2.1.4 adds persisted node metadata fields for failure_domain, provider, region, and rack. These fields are available for placement evolution and audit. They do not constitute a Byzantine proof-of-storage or a formal geographic durability guarantee.

Explicit non-claims

v2.1.4 does not implement controller HA/multi-writer consensus, Byzantine proof-of-storage, on-chain discovery, erasure coding, or a proof that independent failure domains are always available. A single active controller remains the supported topology.

Recommended rollout

  1. Upgrade controller code and schema while preserving the existing SQLite database backup.
  2. Start the v2.1.4 controller in observe with HB_REPL_UNPIN_ENABLED=0.
  3. Upgrade replica agents with HB_REPL_ALLOW_UNPIN=0.
  4. Validate continuous heartbeat and Kubo capacity reporting.
  5. Onboard the second committable node and validate real M=2 behavior.
  6. Move to pin-only and test a canary CID before broad PIN scheduling.
  7. Only after release/refcount/recovery tests pass in production, explicitly enable both UNPIN gates for a controlled canary release.

Never expose the private Kubo API (5011) publicly. Agents must use the local Kubo API only.

Assets 3
Loading