hyperledger-labs / fabric-smart-client / 41410905971

92%

Build Type

push

github

Committed by web-flow

Commit Message

feat(storage): version every FSC store and add node migrate (#2425)

Every FSC store (KVS, binding, signer info, audit info, endorse tx, metadata, envelope and vault) implements migrate.Versioned on SQLite, Postgres and memory. Step 1 runs the CREATE TABLE IF NOT EXISTS statements CreateSchema runs. On the KVS and the vault it also runs the existing layout probes inside the step's transaction, through new ReorderPrimaryKeyTx variants and, on the Postgres vault, convertTextKeysTx. Each probe reads the layout first and takes ACCESS EXCLUSIVE only on a table it has to convert, so step 1 on a v0.27 database records version 1 without blocking older replicas. CreateSchema on these stores is unchanged and records no version.

With skipCreateTable: true an FSC store with pending steps now fails to open with migrate.ErrPending, naming the store; the error tells the operator to run `node migrate`.

While a store runs schema steps or waits for the migration lock, migrate.InProgress reports true and /readyz answers 503 with reason "migrating". This covers a channel's vault, which migrates when the channel opens, after the web listener has started; /healthz stays 200 meanwhile.

`node migrate` installs every SDK without starting it and opens every store the SDKs register, running their pending steps; `--check` runs nothing and exits non-zero while a step is pending. pkg/node gets StoreOpener and Node.Migrate. The view SDK registers one StoreOpener that calls every entry of the dig value group fsc-store-openers, so an SDK nested in another reaches it. The view SDK provides an entry for its stores, and the Fabric SDK one for its stores and the vault of every channel on a generic network. config.Provider.ForceSkipCreateTable sets skipCreateTable on every persistence for the run. Installing loads identities, so the command needs the node's configuration and key material.

Tests:
- Catalog diffs on SQLite and Postgres show that step 1 on an empty database creates the s... (continued)

Coverage Stats

209 of 233 new or added lines in 24 files covered. (89.7%)

5 existing lines in 3 files now uncovered.

30710 of 33540 relevant lines covered (91.56%)

1287.94 hits per line