The indexer image installed @subql/node and the @polkadot -> @pezkuwi aliases
with an unlocked npm install, and the build cache kept the layer from the day
it was first built: the indexers ran @pezkuwi/wasm-crypto 7.5.17/7.5.18, the
build whose ed25519 and secp256k1 are stubs, plus vm2 3.9.19 (the mapping
sandbox, 54 advisories). No lockfile meant Dependabot never saw that tree.
- docker/indexer holds the image's package.json and package-lock.json:
@subql/node 6.4.6 as today, @pezkuwi 16.5.39 / 14.0.26 / 7.5.21, overrides
for vm2 3.12.2, uuid 11 and package-json 7 (got 11). The Dockerfile runs
npm ci on Node 24 (Node 20 is out of support).
- The pruned-state patch stops the build when a pattern matches nothing,
instead of reporting success with the fallback missing.
- CI builds the indexer image and runs the crypto smoke test inside it
through the @polkadot aliases, the way subql-node loads them.
- The root yarn.lock, which the host installs as root on every deploy to run
subql codegen/build, moves the advisory ranges forward with targeted
resolutions and @pezkuwi/braces. Left: elliptic (low, via ethers 5 in
@subql/cli) and sprintf-js (moderate, via jest's argparse 1); neither has
a fixed release.
When SubQuery restarts and hits pruned blocks, events are empty (from patch 4).
This caused two downstream crashes:
- getExtrinsicSuccess(undefined) → TypeError on findIndex
- wrapBlock with events.toArray() → TypeError on undefined
Patches 5-6 handle undefined events gracefully.
When SubQuery restarts and the last processed block's state has been
pruned by the RPC node, fetchEventsRange would crash the indexer in a
loop. Now returns empty events for pruned blocks so the indexer can
skip forward gracefully.
The _getBlockRegistryViaHash method in @polkadot/api calls
getRuntimeVersion.raw(header.parentHash) which fails for genesis
and other pruned blocks. Add fallback to current runtime version.
Substrate nodes prune historical state by default (~256 blocks).
When SubQuery restarts and tries to fetch runtime version for old
blocks, it crashes with "State already discarded". This patch adds
a fallback to current runtime version when historical state is
unavailable, preventing crash loops on pruned RPC endpoints.
Also removes --pull from deploy workflow to avoid unnecessary
full rebuilds on every push.
Satoshi Qazi Muhammed