'Error: memory allocation failed, out of memory' from many msg!/format! calls — the 32KB bump heap never frees

Symptom: a program with many msg!("var: {:?}", var) statements eventually aborts with Program log: Error: memory allocation failed, out of memory. The string does not appear in the validator source you grepped, because it comes from the on-chain allocator path in the SBF runtime, not from invoke_context.rs.

Cause: format! (and therefore msg! with formatting arguments) allocates on the program heap. The default heap is 32KB and the default global allocator is a bump allocator with no free and no realloc — every allocation is permanent for the duration of the instruction. Enough formatted logs and you run out, regardless of compute budget.

What to do: - Request a bigger heap from the client instead of patching the validator: add ComputeBudgetInstruction::request_heap_frame(bytes) to the transaction. The value must be a multiple of 1024 and, as of current Solana/Agave, can go up to 256KB. Editing HEAP_LENGTH in a local validator build does not help, because the effective limit comes from the compute budget, not the program constant. - Log less: prefer msg!("{}", x) on the specific scalar fields you need over {:?} on a whole struct, and use sol_log_data / raw byte logging for hot paths. Formatting also costs compute units, so this helps twice. - For genuinely allocation-heavy programs, install a custom #[global_allocator] that can reuse memory; Solana has published example bump/heap implementations.

Note the failure is instruction-scoped: each instruction gets a fresh heap, so splitting work across instructions also relieves pressure.

Source: adapted from a Stack Exchange answer by McBain (question by Maxim Schmidt), https://solana.stackexchange.com/questions/15904/too-many-debug-prints-causes-out-of-memory — CC BY-SA 4.0