Repository navigation
DKG v10.0.27: faster and steadier when your node is busy
DKG v10.0.27 makes your node faster and steadier when it has a lot to do. It gets back to work quickly after a restart, answers queries sooner, and keeps your agents' shares and publications moving.
npm: @origintrail-official/dkg@10.0.27 — dist-tags latest, testnet, mainnet
What you'll notice
- Back to work fast after a restart. A node catches up on each context graph once and then settles, and an interrupted recovery continues where it stopped.
- Quicker answers. Queries that ask for a limited number of results stop as soon as they have them, and reads across many verified context graphs skip repeated checks.
- Sharing and publishing keep moving. They get store capacity of their own, so other work on a busy node doesn't hold them up, and they stay fast however much history builds up.
- Collaborators stay in step. When a context graph's owner restarts, its members get the change they missed within seconds, and private publishing stays steady while other agents update their keys.
- Your data is safer. Nodes that host shared memory for other members save it crash-safely, and a publication confirmed on chain finishes its catalog step even if the node couldn't store it right away.
Before you update
- Check
config.jsonfor aqueueRetentionblock. This release reads it, so keep cleanupoffon production nodes. - If you call the API, page or narrow a public query whose response is above 10 MiB (it now gets HTTP 413), and retry a share or publish answered with 503 and
Retry-After. - Back up your node as usual.
dkg rollbacktakes it back to 10.0.26 with its data, and nothing changes on chain.
A new log line every five minutes, async_admission_usage, shows where sharing and publishing spend their time. Every change is in the changelog.
Full changelog: v10.0.26...v10.0.27