The legacy scale test cluster setup script waits for each component serially which means it will abort early if the components that are supposed to come up relatively quickly (i.e. central and central-db) fail to become healthy soon. Then it will wait for the slower ones. As @davdhacs commented this has the benefit of freeing up resources (a large cluster is relatively expensive) early if something is badly broken.
OTOH roxie (IIRC) waits for all components in parallel and has only one --*-wait flag to control the overall timeout. This timeout needs to be large if we want roxie to also wait for the slower components. But if it's large, and central is broken beyond repair, then we'll be wasting time and thus money.
I think if we want to keep the wait logic within roxie (and avoid adding custom bash wait functions everywhere), we need to introduce some heuristics or finer-grained (per-component?) timeouts.
The legacy scale test cluster setup script waits for each component serially which means it will abort early if the components that are supposed to come up relatively quickly (i.e. central and central-db) fail to become healthy soon. Then it will wait for the slower ones. As @davdhacs commented this has the benefit of freeing up resources (a large cluster is relatively expensive) early if something is badly broken.
OTOH roxie (IIRC) waits for all components in parallel and has only one
--*-waitflag to control the overall timeout. This timeout needs to be large if we want roxie to also wait for the slower components. But if it's large, and central is broken beyond repair, then we'll be wasting time and thus money.I think if we want to keep the wait logic within roxie (and avoid adding custom bash wait functions everywhere), we need to introduce some heuristics or finer-grained (per-component?) timeouts.