provider changed before inventory refresh completed

Error message

provider changed before inventory refresh completed

What it means

A background provider-inventory refresh captured a refresh_identity tied to a provider id, then re-fetched the session's provider and compared names; the session had switched to a different provider before the refresh finished. The refresh aborts so models fetched from the old provider are never stored under the new provider's identity — a consistency guard against a provider-switch race, not a data-loss failure.

Source

Thrown at crates/goose/src/acp/server/dispatch.rs:231

                                let agent_bg = agent.clone();
                                let cx_bg = cx.clone();
                                let session_id_bg = session_id.clone();
                                tokio::spawn(async move {
                                    let refresh_identity = refresh_job.identity;
                                    let refresh_provider_id = refresh_job.provider_id;
                                    let mut refresh_guard =
                                        agent_bg.provider_inventory.refresh_guard(&refresh_identity);
                                    let provider_result: Result<Arc<dyn Provider>> =
                                        AssertUnwindSafe(async {
                                            let session_agent =
                                                agent_bg.get_session_agent(&session_id_bg.0).await?;
                                            let provider = session_agent
                                                .provider()
                                                .await
                                                .map_err(|e| anyhow::anyhow!(e.to_string()))?;
                                            let provider_name = provider.get_name().to_string();
                                            if provider_name != refresh_provider_id {
                                                return Err(anyhow::anyhow!(
                                                    "provider changed before inventory refresh completed"
                                                ));
                                            }
                                            Ok(provider)
                                        })
                                        .catch_unwind()
                                .await
                                .map_err(|_| {
                                    anyhow::anyhow!("provider inventory refresh task panicked")
                                })
                                .and_then(|result| result);

                                let fetch_result = match provider_result {
                                    Ok(provider) => {
                                        match ensure_refresh_identity_current(
                                            &refresh_provider_id,
                                            &refresh_identity,
                                        )

View on GitHub (pinned to 3810898a74)

Solutions

Treat it as benign: switch providers again or wait — the next refresh for the new provider runs with a fresh identity
Avoid changing providers repeatedly in quick succession while a model list is loading
If it persists, check that provider get_name() is stable across calls (custom providers returning dynamic names would trip this guard)

Defensive patterns

Strategy: retry

Try / catch

// guard result is consumed by the refresh task itself:
if let Err(e) = refresh_result {
    if e.to_string().contains("provider changed before") {
        tracing::debug!(%e, "stale refresh skipped; new refresh already scheduled");
    } else {
        tracing::warn!(%e, "inventory refresh failed");
    }
}

Prevention

Avoid switching providers while a model-list refresh is in flight
Treat this guard as informational — the new provider's refresh carries fresh data

When it happens

Trigger: The user changes the model/provider in the UI while a recommended-models refresh for the previous provider is still in flight; the check 'provider_name != refresh_provider_id' then trips and the task errors out.

Common situations: Rapid provider switching in the desktop app while background refreshes are pending; flaky network making the original refresh slow enough that the user switches providers meanwhile.

Related errors

AI-assisted analysis of aaif-goose/goose@3810898a74 (2026-08-16). Data as JSON: /api/errors/4fc158a6f69dbb40. Report an issue: GitHub.