matlab-set-up-worker-state

Use parallel.pool.Constant to send data to workers once for the pool lifetime instead of re-broadcasting every loop. For file-based data, use the build-function form parallel.pool.Constant(@() load(file).var) so each worker loads directly from disk.

Use createConnectionForPool from Database Toolbox, which returns a parallel.pool.Constant holding a per-worker connection. Connections are torn down automatically when the Constant goes out of scope or the pool shuts down.

Objects like database connections, loaded libraries, and file handles cannot be serialized and sent from the client to workers. Instead, use parallel.pool.Constant with a build function so each worker constructs the resource locally.

Prefer parallel.pool.Constant over spmd for worker state setup. spmd setup has no cleanup guarantee if parfor errors, makes the pool fragile to worker disconnection, and produces Composite variables unusable inside parfor.

Use parfevalOnAll(pool, @addpath, 0, p) to add paths on all workers of an existing pool. Never delete and recreate the pool just to change paths, since pool startup can take 30 seconds or more.

Yes, parallel.pool.Constant and parfevalOnAll work on thread pools. However, thread workers share the client process, so AdditionalPaths and EnvironmentVariables do not apply; modify the client environment before the parfor instead.