Slow perceived response time due to excessive storage writes during streaming

[taetaetae]

Description

During streaming responses, OpenCode writes to storage (JSON files) on every reasoning-delta event. With fast streaming APIs, this results in hundreds of file writes per second.

Impact

• High I/O overhead from frequent file system operations
• Increased latency between receiving content and displaying it
• Slower perceived response time compared to native CLI tools (e.g., kiro-cli)

Root Cause

In processor.ts, every delta triggers a storage write:

case "reasoning-delta":
  if (part.text) await Session.updatePart({ part, delta: value.text })

Session.updatePart calls Storage.write which writes to the file system on every call.

Proposed Solution

Throttle storage writes during streaming:

• Accumulate content in memory
• Only flush to storage every 50ms
• Always flush on reasoning-end

Environment

• OpenCode version: latest
• Tested with Kiro API streaming responses

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• High CPU usage (100%+) during LLM streaming in long sessions due to O(n) text buffer rendering #6172: High CPU usage (100%+) during LLM streaming in long sessions due to O(n) text buffer rendering

Feel free to ignore if none of these address your specific case.

[taetaetae]

Thanks for flagging #6172! After reviewing it, I believe these are different issues addressing different bottlenecks:

Aspect	#6172 (OpenTUI)	#11329 (This issue)
Root cause	O(n) text buffer rendering in Zig renderer	Excessive file I/O on every streaming delta
Hot path	textBufferViewMeasureForDimensions → mmap	Session.updatePart → Storage.write → Bun.write
Layer	TUI rendering (OpenTUI native)	Application logic (TypeScript)
Fix location	OpenTUI Zig code	processor.ts

#6172 addresses CPU usage from re-rendering the entire text buffer on every update.

This issue addresses I/O overhead from writing to disk on every streaming delta in processor.ts:

case "reasoning-delta":
  await Session.updatePart({ part, delta: value.text })  // File write every delta

Both optimizations are complementary - fixing one doesn't fix the other. The PR #11328 addresses this specific storage I/O bottleneck.

[buildkloe]

Confirming significant disk I/O impact

Experiencing 11.2 MB/s sustained writes from OpenCode (desktop and CLI) during normal usage or idle state - more than any other process on the system. This is visible in macOS Activity Monitor as continuous disk activity even no operation at all.

I'm a bit worry about the impact:
SSD wear from constant writes during every session
Competes with other disk I/O (spotlight indexing)
Battery drain on laptops
Would love to see FIX/PR #11328 prioritized.

Thank you!

[github-actions]

To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.