Skip to content

Agent Runtime Limits ​

Use this reference when a job produces incomplete output, a transfer exceeds endpoint capacity, or an Agent requires recovery.

Process output ​

Blocking scripts and workflow commands retain up to 1 MiB each of stdout and stderr by default. Extra output is discarded and cannot be recovered. Write large diagnostics to a file and collect it as an artifact.

Keep structured JSON output within the limit; truncated JSON cannot be parsed reliably. Docker Compose configuration discovery can also fail when its configuration exceeds this limit.

To change the retained output limit in local Agent configuration:

toml
[execution]
output_limit_bytes = 1048576

The value applies to newly started processes.

Process cancellation ​

Aborting work requests cancellation but cannot undo changes already made. Detached processes and work accepted by Windows servicing or an external system may continue. Non-blocking Windows scripts are not stopped by job cancellation.

On Linux, the default graceful-stop period is five seconds before forced termination. Configure it locally when a script needs a different shutdown period:

toml
[execution]
termination_grace_seconds = 5

Artifact downloads ​

Downloads require HTTPS and enough free disk space for the file plus a default 256 MiB reserve. Invalid or incomplete downloads are not made available for use. Check connectivity, disk space, and the job timeout when a transfer fails.

Archive extraction ​

Package ZIPs and WinPE driver bundles must fit the endpoint's extraction limits. Defaults allow 100,000 entries, 50 GiB of expanded content, a path depth of 32, and a 100:1 compression ratio for entries of at least 1 MiB. Archives must not contain unsafe paths or symbolic links.

Retrieved files and directories ​

Retrieved files and directory archives are limited to 15 GiB by default. Directory retrieval skips symbolic links. Split larger collections or adjust the limits after checking storage and transfer capacity.

Configuring artifact limits ​

These local configuration values control the supported limits:

toml
[artifacts]
max_upload_bytes = 16106127360
max_archive_entry_count = 100000
max_archive_expanded_bytes = 53687091200
max_archive_entry_bytes = 53687091200
max_archive_path_depth = 32
max_archive_compression_ratio = 100
download_free_space_reserve_bytes = 268435456
archive_free_space_reserve_bytes = 268435456

Restart the Agent after changing artifact limits. Larger limits require more disk space and potentially longer job timeouts.

Instruction replay synchronization ​

After an Agent update or reinstall, Ordyn automatically clears a recovery block once the release outcome is confirmed and no other command outcomes remain unknown.

Select Recover replay protection from the endpoint lifecycle menu or sidebar context menu with Endpoint Operator to recover blocked instructions and interrupted jobs. You can request recovery while the endpoint is offline; it completes automatically when the agent reconnects and confirms synchronization. Affected old jobs close with an unknown outcome and are not rerun. An external change may already have happened.

Avoid restoring enrolled Agent state from virtual-machine snapshots or old filesystem backups. Re-enroll the device before resuming work when replaying an earlier action would be unacceptable.

Instruction supervision and restart recovery ​

An Agent restart can interrupt running work. Some preparation steps recover automatically; inspect other failed tasks before rerunning them. An endpoint may remain busy while interrupted work is stopping.

For maintenance that interrupts Agent connectivity, see durable result spooling.

Local state persistence ​

If the Agent reports corrupt state, stop its service before attempting repair. Inspect the retained recovery points:

bash
ordyn-agent repair-state

Only select a valid recovery point for the same endpoint when losing changes made after that point is acceptable:

bash
ordyn-agent repair-state --generation 42

Restart the service after successful repair. If the command refuses recovery or no valid point remains, use endpoint credential repair where applicable or contact support before proceeding.

Agent downgrade ​

Reusing an existing Agent state directory with an older Agent version is unsupported. Restore the matching version or enroll with a new state directory.