Skip to content

Sdd/parallel reasoning agent - #96

Closed
bxyu-nvidia wants to merge 3 commits into
mainfrom
sdd/parallel-reasoning-agent
Closed

Sdd/parallel reasoning agent#96
bxyu-nvidia wants to merge 3 commits into
mainfrom
sdd/parallel-reasoning-agent

Conversation

@bxyu-nvidia

Copy link
Copy Markdown
Contributor

No description provided.

@copy-pr-bot

copy-pr-bot Bot commented Sep 25, 2025

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@bxyu-nvidia bxyu-nvidia closed this Nov 7, 2025
@bxyu-nvidia
bxyu-nvidia deleted the sdd/parallel-reasoning-agent branch November 7, 2025 21:40
hemildesai added a commit that referenced this pull request Jul 15, 2026
…am drops

OpenSandbox exec streams a command's output over one long-lived SSE connection
through the server proxy. When that connection crosses a load balancer with a
TCP idle/stream-duration cap (e.g. an AWS NLB fronting an istio east-west
gateway, default 350s), a command that runs a while gets its stream silently
reaped and the client hangs (~4h) since the SSE read timeout is disabled.

Add OpenSandboxOperationConfig.background_exec (default False): submit the
command in the background, then issue short get_command_status / get_background_
command_logs polls instead of holding one stream open. No single request is
long enough to be reaped, so it is robust across LB idle/duration limits. Also
default command_retries 0->3 in the shipped config (recovers RST-family drops)
and enable background_exec there since OpenSandbox exec always traverses the
server proxy. Root cause confirmed via OpenSandbox issues #1277/#507/#96 and a
mesh trace (sbx east-west NLB, 350s idle, no override).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant