Is Weaviate Down?

If Prismix shows Weaviate as "Operational" but your queries or imports are failing:

• Hybrid search BM25 returning 0 results (tokenizer language mismatch) — Weaviate's BM25 keyword search uses a tokenizer configured per class at schema creation time. If your class uses the word tokenizer but your text contains non-English characters or CJK, BM25 may tokenize poorly and return 0 matches. Rebuild the schema with the field tokenizer or use vector-only search for multilingual content.
• Generative module rate limit from the underlying LLM provider — Weaviate's generative-openai and generative-anthropic modules pass requests to the respective LLM API. Rate limit errors from the LLM provider surface as 500 errors from Weaviate's generative module. Monitor your OpenAI or Anthropic rate limits separately, or configure multiple API keys with load balancing in your Weaviate module settings.
• Schema migration blocking writes — adding a new property to an existing Weaviate class triggers a schema migration that temporarily blocks writes to that class. During the migration, upsert requests return 409 Conflict. Design your schema upfront to avoid frequent migrations in production; if migration is unavoidable, schedule it during a low-traffic window.
• Multi-tenant tenant not found after creation — in Weaviate's multi-tenancy mode, tenants must be explicitly activated after creation. A newly added tenant starts in COLD (inactive) state by default. Call the activate tenant endpoint before attempting to read or write, or create tenants with activityStatus: "ACTIVE" in the create request.
• gRPC connection timeout during streaming batch import — Weaviate's gRPC batch import streams data over a long-lived connection. Firewalls or proxies with aggressive connection timeouts (under 60 seconds) will kill the stream mid-import. Set grpc.keepalive_timeout_ms in your client and ensure your network allows long-lived TCP connections on port 50051.
• WCS cluster memory spike during re-indexing — Weaviate Cloud Services clusters may experience high memory utilization when rebuilding HNSW indexes after a large bulk import. During this period, query latency increases and the cluster may appear degraded on status.weaviate.io. Monitor the cluster's memory metrics in the WCS console; re-indexing completes within minutes to hours depending on collection size.