Give a self-hosted SigNoz at least 4 GB of memory before debugging failing pods. When data stops appearing, start at the collector and work backward toward the app, not the reverse. Treat the ClickHouse PVCs as your data field - back them up before any uninstall, and only delete them when you truly mean to wipe history.

Context: A SigNoz self-host deployment guide documents the resource and uninstall traps. The stack brings ClickHouse, ClickHouse Keeper, Postgres, and an OTel collector, so budget at least 4 GB of Docker memory or components fail to start. When no metrics appear, the checklist is: confirm the OTel collector is running, verify services can reach it, read the collector logs, and check the agent configuration. Uninstalling via helm leaves PVCs behind; deleting them destroys the data.