zylon-ai/private-gpt · error · Errors.Overloaded

OVERLOADED_CONDENSATION_ERROR

OVERLOADED_CONDENSATION_ERROR

Error message

Condensation strategy timed out after {timeout_seconds} seconds

What it means

Wrapped as Errors.Overloaded with event_code OVERLOADED_CONDENSATION_ERROR when the condensation strategy's async get_memory() call does not finish within timeout_seconds (enforced via asyncio.wait_for). It signals the backend (typically an LLM summarization call) is too slow or saturated, not that the history is malformed.

Source

Thrown at private_gpt/components/chat/processors/chat_history/memory/tldr_processor.py:195

async def _apply_condensation_strategy(
    chat_history: list[ChatMessage],
    condense_strategy: BaseMemoryStrategy,
    max_length: int,
    timeout_seconds: int | None = None,
    **kwargs: dict[str, Any],
) -> list[ChatMessage]:
    """Apply the specified condensation strategy to chat history."""
    try:
        coro = condense_strategy.get_memory(
            chat_history=chat_history, max_length=max_length, **kwargs
        )
        if timeout_seconds is None:
            return await coro

        return await asyncio.wait_for(coro, timeout=timeout_seconds)
    except TimeoutError as e:
        raise Errors.Overloaded(
            f"Condensation strategy timed out after {timeout_seconds} seconds",
            event_code=Errors.Codes.OVERLOADED_CONDENSATION_ERROR,
        ) from e

View on GitHub (pinned to 4a030776a3)

Solutions

1. Increase timeout_seconds to comfortably exceed p99 LLM latency for summarization
2. Pass timeout_seconds=None to disable the timeout when condensation is trusted to finish
3. Reduce the history size fed to the summarizer (fewer messages = faster call)
4. Address backend saturation: more capacity, faster model, or load shedding upstream

Example fix

# before
history = await apply_condense_strategy(
    condense_strategy, chat_history, max_length,
    timeout_seconds=5,
)

# after
history = await apply_condense_strategy(
    condense_strategy, chat_history, max_length,
    timeout_seconds=120,
)

try:
    history = await apply_condense_strategy(strategy, chat_history, max_length, timeout_seconds=60)
except Errors.Overloaded as e:
    if e.event_code == Errors.Codes.OVERLOADED_CONDENSATION_ERROR:
        history = await apply_condense_strategy(strategy, chat_history, max_length, timeout_seconds=180)
    else:
        raise

When it happens

Trigger: Calling the strategy wrapper with a finite timeout_seconds while condense_strategy.get_memory() (LLM-backed summarization) takes longer; a slow or overloaded LLM endpoint; a huge conversation being summarized in one call.

Common situations: Self-hosted LLM under load; network latency spikes; timeout_seconds set tighter than typical LLM latency; retry storms amplifying backend slowness.

Understand the failure class

• Timeouts: ETIMEDOUT, deadlines, and hung requests — what actually expires when a request times out.

Related errors

• Server did not become ready within {_HEALTH_TIMEOUT}s
• Unknown condense strategy: {strategy}
• No user messages found in the chat history.
• Maximum number of iterations for condensing exceeded.
• The last user message exceeds the maximum length allowed.

AI-assisted analysis of zylon-ai/private-gpt@4a030776a3 (2026-08-15). Data as JSON: /api/errors/e62b4019afb01a5e. Report an issue: GitHub.