infiniflow/ragflow · warning · RuntimeError

SSH execution output exceeded {self.max_output_bytes} bytes.

Error message

SSH execution output exceeded {self.max_output_bytes} bytes.

What it means

Raised as RuntimeError by _validate_output_size after a command completes, when encoded stdout+stderr combined exceed self.max_output_bytes (default 1 MiB, configurable at initialize). It fires after execution succeeded — the process ran, output was captured, then rejected — so the run's output is lost when it raises. This guards the agent pipeline from multi-megabyte dumps entering structured-result extraction.

Source

Thrown at agent/sandbox/providers/ssh.py:607

            if time.time() > deadline:
                channel.close()
                raise TimeoutError(f"Execution timed out after {timeout} seconds")
            time.sleep(0.1)

        while channel.recv_ready():
            stdout_chunks.append(channel.recv(65536))
        while channel.recv_stderr_ready():
            stderr_chunks.append(channel.recv_stderr(65536))

        exit_code = channel.recv_exit_status()
        stdout = b"".join(stdout_chunks).decode("utf-8", errors="replace")
        stderr = b"".join(stderr_chunks).decode("utf-8", errors="replace")
        return stdout, stderr, exit_code

    def _validate_output_size(self, stdout: str, stderr: str) -> None:
        output_size = len((stdout or "").encode("utf-8")) + len((stderr or "").encode("utf-8"))
        if output_size > self.max_output_bytes:
            raise RuntimeError(f"SSH execution output exceeded {self.max_output_bytes} bytes.")

    def _collect_artifacts(
        self,
        sftp: paramiko.SFTPClient,
        artifacts_dir: str,
    ) -> list[dict[str, Any]]:
        artifacts: list[dict[str, Any]] = []
        self._collect_artifacts_recursive(sftp, artifacts_dir, "", artifacts)
        return artifacts

    def _collect_artifacts_recursive(
        self,
        sftp: paramiko.SFTPClient,
        current_dir: str,
        relative_dir: str,
        artifacts: list[dict[str, Any]],
    ) -> None:
        try:

View on GitHub (pinned to 554fb1133a)

Solutions

Reduce output in the sandboxed code: print head()/summaries, write bulk data to files instead of stdout
Raise the limit at initialize: config {'max_output_bytes': 4*1024*1024} if the pipeline can absorb it
Capture stderr separately in your code so warnings don't inflate combined size

Example fix

# before
provider.execute_code(inst, "print(open('big.csv').read())", "python")

# after
provider.execute_code(inst, "print(open('big.csv').read()[:2000])", "python")

Defensive patterns

Strategy: validation

Validate before calling

estimated = len(stdout_text.encode()) + len(stderr_text.encode())
if estimated > provider.max_output_bytes:
    raise RuntimeError("output will exceed provider limit; print a summary instead")

Try / catch

try:
    result = provider.execute_code(instance_id, code, language)
except RuntimeError as e:
    if "exceeded" in str(e) and "bytes" in str(e):
        code = code.replace("print(df)", "print(df.head())")
        result = provider.execute_code(instance_id, code, language)

Prevention

Default sandboxed code to previews: head(), first N lines, truncated strings
Size max_output_bytes deliberately at initialize based on what downstream LLM context can absorb
Route bulk data to files/artifacts, not stdout

When it happens

Trigger: Code that prints large data (dumping a DataFrame, cat-ing a big file, verbose logs); combining stdout and stderr just over the limit; lowering max_output_bytes in config below typical output size; the structured-result marker plus large payload.

Common situations: LLM code agents printing entire datasets instead of summaries; debugging prints left in generated code; default 1 MiB too small for legitimate CSV previews.

Related errors

AI-assisted analysis of infiniflow/ragflow@554fb1133a (2026-08-15). Data as JSON: /api/errors/be3e6b2136999a22. Report an issue: GitHub.