CacheStore cannot be used on the worker node! eviction was u

Error message

CacheStore cannot be used on the worker node! eviction was used.

What it means

CacheStore::eviction triggers cache eviction policy execution. On worker nodes it is a deliberate panic! stub (cache_rocksstore.rs:2084): eviction is a master-side responsibility and worker cache stores never run it. The panic indicates eviction was requested on the wrong node.

Source

Thrown at rust/cubestore/cubestore/src/cachestore/cache_rocksstore.rs:2084

        _timeout: u64,
    ) -> Result<Option<QueueResultResponse>, CubeError> {
        panic!("CacheStore cannot be used on the worker node! queue_result_blocking was used.")
    }

    async fn queue_merge_extra(&self, _key: QueueKey, _payload: String) -> Result<(), CubeError> {
        panic!("CacheStore cannot be used on the worker node! queue_merge_extra was used.")
    }

    async fn compaction(&self) -> Result<(), CubeError> {
        panic!("CacheStore cannot be used on the worker node! compaction was used.")
    }

    async fn info(&self) -> Result<CachestoreInfo, CubeError> {
        panic!("CacheStore cannot be used on the worker node! info was used.")
    }

    async fn eviction(&self) -> Result<EvictionResult, CubeError> {
        panic!("CacheStore cannot be used on the worker node! eviction was used.")
    }

    async fn persist(&self) -> Result<(), CubeError> {
        panic!("CacheStore cannot be used on the worker node! persist was used.")
    }

    async fn healthcheck(&self) -> Result<(), CubeError> {
        panic!("CacheStore cannot be used on the worker node! healthcheck was used.")
    }

    async fn rocksdb_properties(&self) -> Result<Vec<RocksPropertyRow>, CubeError> {
        panic!("CacheStore cannot be used on the worker node! rocksdb_properties was used.")
    }

    async fn wipe(&self) -> Result<(), CubeError> {
        panic!("CacheStore cannot be used on the worker node! wipe was used.")
    }
}

View on GitHub (pinned to 7d981676b3)

Solutions

Run eviction only on the master node's store
Point management commands to the master endpoint
Add an is-master assertion before invoking eviction in shared code

Example fix

// before
store.eviction().await?; // runs everywhere
// after
if is_master { master_store.eviction().await?; }

Defensive patterns

Strategy: validation

Validate before calling

if !is_master { return Err(CubeError::internal("eviction is master-only")); }
master_client.eviction().await?;

Type guard

fn supports_eviction(store: &dyn CacheStore) -> bool { node_role == NodeRole::Master }

Try / catch

std::panic::catch_unwind(AssertUnwindSafe(|| master_client.eviction())).map_err(|_| CubeError::internal("eviction called on worker node"))?;

Prevention

Run eviction loops only on the master
Gate eviction triggers with role assertions
Review topology code after failover/promotion
Monitor for panic-rate spikes on worker nodes

When it happens

Trigger: Calling CacheStore::eviction() on a worker-node CacheStore (cache_rocksstore.rs:2084), e.g. from an eviction loop or admin command routed to a worker.

Common situations: Eviction/management loops misconfigured to run cluster-wide; topology changes where a worker inherited master-era scheduler code.

Related errors

AI-assisted analysis of cube-js/cube@7d981676b3 (2026-09-02). Data as JSON: /api/errors/0e008437478a990c. Report an issue: GitHub.