`java.lang.InternalError: a fault occurred in an unsafe memory access operation` (Zilla 0.9.173)

[nikhil-cli]

Zilla crashes while serving a request routed through the Kafka cache.
The engine terminates after throwing:

java.lang.InternalError: a fault occurred in an unsafe memory access operation

This originates from the Kafka cache internals (KafkaCacheFile.readBytes).
Immediately afterwards, many repeated messages appear:

java.lang.IllegalStateException: Engine worker usage is non-zero: <various negative or small values>

Finally the engine stops:

ENGINE_STOPPED Engine Stopped.

This causes the entire Zilla container to crash.

---

To Reproduce

1. Deploy Zilla 0.9.173 with Kafka cache enabled.
2. Make a request that triggers a Kafka fetch via cache (e.g., SSE route consuming from Kafka).
3. Zilla attempts to read cached data.
4. The engine crashes with an internal memory fault.

---

Expected Behavior

Zilla should safely read from the Kafka cache without triggering unsafe memory faults or crashing the entire engine.

---

Crash Logs

java.lang.InternalError: a fault occurred in an unsafe memory access operation
    at KafkaCacheFile.readBytes(...)
    at KafkaCacheCursor.next(...)
    at KafkaCacheClientFetchStream.doClientReplyDataIfNecessary(...)
    ...
Caused by: java.lang.InternalError: a fault occurred in an unsafe memory access operation

Followed by dozens of:

java.lang.IllegalStateException: Engine worker usage is non-zero: -263
java.lang.IllegalStateException: Engine worker usage is non-zero: -164
java.lang.IllegalStateException: Engine worker usage is non-zero: 1
...
ENGINE_STOPPED Engine Stopped.

---

Zilla Environment

• Zilla version: 0.9.173
• Running inside Kubernetes
• Using Kafka cache (kafka_cache_server / kafka_cache_client)
• Route: SSE → Kafka via cache

---

Kafka Environment

• Standard Kafka cluster
• Topic being consumed contains time-series style data
• No unusual Kafka errors seen outside Zilla

---

Additional Context

• Crash happens inside unsafe memory access within the Kafka cache, not the TCP/Kafka client.
• This appears to be low-level memory corruption or an edge-case triggered by reading cache segments.
• After the crash, workers report negative usage counts, suggesting resource-accounting corruption after the fault.
• Container restarts automatically after engine shutdown.