fix(node): don't spawn Frontier-RPC pool validation tasks as essential

[Kailai-Wang]

Problem

Every v0.9.26 node image crashes immediately on heima — even on a fresh genesis DB at #0:

[Parachain] Essential task `transaction-pool-task-0` failed. Shutting down service.
[Parachain] Essential task `transaction-pool-task-1` failed. Shutting down service.
Error: Service(Other("Essential task failed."))

This is why production has never been able to move off v0.9.25-01. It is not a host-function or runtime-version issue (the on-chain runtime imports zero moonbeam host functions); it is a node-side wiring bug.

Root cause

The standard-frontier migration (#3841, commit 305767d2d) added a secondary transaction pool, used only to expose a low-level sc_transaction_pool::Pool<FullChainApi> (graph) to the standard Frontier txpool/graph RPC (which requires the inner pool type, unlike the old fork that accepted the high-level handle):

let chain_api = Arc::new(sc_transaction_pool::FullChainApi::new(
    client.clone(), None, &task_manager.spawn_essential_handle(),  // <-- essential
));

FullChainApi::new spawns two transaction-pool-task-{0,1} validation tasks via that spawner. Their loop returns when the validation channel closes, and they resolve at startup. Because they were spawned with the essential handle, an essential task ending (not even panicking) triggers task_manager's "Essential task failed. Shutting down service." — tearing the whole node down. (No panic/backtrace ever appears precisely because it's a normal task completion, not a panic.)

Fix

Spawn that secondary RPC-only pool's tasks non-essentially via a small NonEssentialSpawner adapter that implements SpawnEssentialNamed by forwarding to the ordinary SpawnNamed methods of a SpawnTaskHandle. The Frontier RPC graph still works; those tasks completing is no longer fatal. The primary transaction pool (and --pool-type) is untouched.

Testing

• Reproduces deterministically on a fresh heima genesis DB with any v0.9.26 build; the crash disappears with this change in principle (validation-task completion no longer fatal).
• Could not fully type-check locally (unrelated macOS rpassword/__errno_location toolchain issue blocks the build on the dev machine before reaching the node crate); relying on CI (Linux) for the compile, and recommend validating the built image against a heima DB snapshot before any production swap.

Note

Keep prod on v0.9.25-01 until an image built from this is verified to start and author on a heima snapshot.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
heima-aa-demo-app	Ignored		Jun 25, 2026 12:43pm