Solana programs run on-chain, so debugging them in the wild can be challenging. To make debugging programs easier, developers can write unit tests that directly test their program's execution via the Solana runtime, or run a local cluster that will allow RPC clients to interact with their program.
During program execution both the runtime and the program log status and error messages.
For information about how to log from a program see the language specific documentation:
When running a local cluster the logs are written to stdout as long as they are
enabled via the
RUST_LOG log mask. From the perspective of program
development it is helpful to focus on just the runtime and program logs and not
the rest of the cluster logs. To focus in on program specific information the
following log mask is recommended:
Log messages coming directly from the program (not the runtime) will be displayed in the form:
Program log: <user defined message>
The amount of information that can be communicated via a transaction error is limited but there are many points of possible failures. The following are possible failure points and information about what errors to expect and where to get more information:
- The BPF loader may fail to parse the program, this should not happen since the
loader has already finalized the program's account data.
InstructionError::InvalidAccountDatawill be returned as part of the transaction error.
- The BPF loader may fail to setup the program's execution environment
InstructionError::Custom(0x0b9f_0001)will be returned as part of the transaction error. "0x0b9f_0001" is the hexadecimal representation of
- The BPF loader may have detected a fatal error during program executions
(things like panics, memory violations, system call errors, etc...)
InstructionError::Custom(0x0b9f_0002)will be returned as part of the transaction error. "0x0b9f_0002" is the hexadecimal representation of
- The program itself may return an error
InstructionError::Custom(<user defined value>)will be returned. The "user defined value" must not conflict with any of the builtin runtime program errors. Programs typically use enumeration types to define error codes starting at zero so they won't conflict.
In the case of
VirtualMachineFailedToRunProgram errors, more information about
the specifics of what failed are written to the program's execution
For example, an access violation involving the stack will look something like this:
BPF program 4uQeVj5tqViQh7yWWGStvkEG1Zmhx6uasJtWCJziofM failed: out of bounds memory store (insn #615), addr 0x200001e38/8
The program can log the remaining number of compute units it will be allowed before program execution is halted. Programs can use these logs to wrap operations they wish to profile.
See compute budget for more information.
The BPF shared object internals can be dumped to a text file to gain more insight into a program's composition and what it may be doing at runtime.
During execution the runtime BPF interpreter can be configured to log a trace message for each BPF instruction executed. This can be very helpful for things like pin-pointing the runtime context leading up to a memory access violation.
The trace logs together with the ELF dump can provide a lot of insight (though the traces produce a lot of information).
To turn on BPF interpreter trace messages in a local cluster configure the
solana_rbpf level in
trace. For example: