`remote:drush sql-dump` truncates large output due to OutputFormatter overhead

[dpagini]

Summary

terminus remote:drush <site.env> -- sql-dump silently truncates output for large databases (~886 MB). The dump ends mid-INSERT with no -- Dump completed marker and exit code 0. Small databases (~1.5 MB) work fine.

This is a regression introduced in Terminus v4 via the upgrade to symfony/console 6.x.

Root Cause

SSHBaseCommand::getOutputCallback() writes SSH output through $output->write($buffer) using the default OUTPUT_NORMAL mode. This routes every chunk of SQL data through OutputFormatter::format(), which regex-parses the data looking for console-style tags (<info>, <error>, etc.).

Symfony Console 6.x introduced a ~27× performance regression in OutputFormatter::formatAndWrap() (see upstream issue). The slowdown creates backpressure through the SSH pipe, throttling the remote drush sql-dump process until it's killed by server-side execution limits—resulting in a silently truncated dump.

Two Issues

1. Performance regression (v4 only)

The symfony/console 6.x formatter is ~27× slower on HTML-rich text. For an 886 MB dump:

	Terminus v3 (Console 5.4)	Terminus v4 (Console 6.4)
Formatter time per chunk	7.8ms	212ms
Projected total	~3.6 min	~99 min
Dump result	✅ Complete (886 MB)	❌ Truncated (~848 KB)

2. Latent data corruption (v3 and v4)

Both versions silently strip Symfony Console style tags from piped output. If any database content contains literal <info>, <error>, <comment>, <question>, or <fg=...> strings, they are removed from the dump:

Input:  INSERT INTO node__body VALUES ('<info>Important notice</info>');
Output: INSERT INTO node__body VALUES ('Important notice');

This is rare in practice but means dumps are not byte-accurate even when they complete.

Suggested Fix

In src/Commands/Remote/SSHBaseCommand.php, the getOutputCallback() method should use OUTPUT_RAW:

private function getOutputCallback()
{
    $output = $this->output();
    $stderr = $this->stderr();

    return function ($type, $buffer) use ($output, $stderr) {
        if (Process::ERR === $type) {
            $stderr->write($buffer);
        } else {
            $output->write($buffer, false, OutputInterface::OUTPUT_RAW);
        }
    };
}

OUTPUT_RAW bypasses OutputFormatter::format() entirely, which is correct for streaming raw command output. The formatter is designed for styled console messages, not for piping data.

This fixes both issues: the performance regression (no formatting overhead) and the data corruption (no tag stripping).

Reproduction

# Large database (fails with v4)
terminus remote:drush <site.env> -- sql-dump --structure-tables-list=cache,cache_* | wc -c

# Compare: same command via Terminus v3 produces ~886MB, v4 produces ~848KB

Environment

• Terminus v4.1.4 (symfony/console v6.4.32)
• Terminus v3.6.1 (symfony/console v5.4.43) — works correctly
• PHP 8.x

Disclaimer: Bug report heavily written by AI...

[namespacebrian]

@dpagini are you using an agentic AI to identify root cause? If so, could you ask your agent to send us a PR with the same change proposed here? :) It would expedite our testing..

[namespacebrian]

Nevermind.. I just realized I can point my claude at this issue ticket and tell it to create the PR :)

(I think recognizing opportunities to use these new tools is one of the most important skills for devs to work on building going forward)

[namespacebrian]

I should mention that we have some reservations/concerns about how this might change the behavior for other customers who expect the current behavior. We have to give it some thought and testing.

[dpagini]

With "the current behavior" being...? Styled output of SSH commands? or something else? I did add a comment to the PR about that, but just curious what the other thoughts might be. Could a flag be passed in the interim, like --raw to control this maybe?