Skip to content

feat: smart provider failover #971

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
pipi-1997 wants to merge 11 commits into from
Closed

feat: smart provider failover #971

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
d98e091
feat: smart provider failover
pipi-1997 Jan 21, 2026
e44a4c0
docs: clarify failover strategies and model array config
pipi-1997 Jan 21, 2026
b71028a
refactor(config): remove unused failover config
pipi-1997 Jan 21, 2026
140fa01
docs: tighten smart failover docs
pipi-1997 Jan 21, 2026
aaeb7c1
fix(failover): parse rate limit reset correctly
pipi-1997 Jan 21, 2026
e3e16b1
test(smart-failover): await toast scheduling
pipi-1997 Jan 21, 2026
e806b63
refactor(smart-failover): address copilot review
pipi-1997 Jan 22, 2026
5d46bb2
docs(smart-failover): clarify enablement and cooldown
pipi-1997 Jan 22, 2026
2fc548e
fix(sisyphus-junior): handle empty model override arrays
pipi-1997 Jan 23, 2026
daa4a98
merge dev into feat/smart-provider-failover
Jan 24, 2026
3324ea0
fix: resolve dev merge type issues~
pipi-1997 Jan 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
184 changes: 169 additions & 15 deletions assets/oh-my-opencode.schema.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,19 @@
"$schema": {
"type": "string"
},
"model": {
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"disabled_mcps": {
"type": "array",
"items": {
Expand Down Expand Up @@ -77,7 +90,8 @@
"delegate-task-retry",
"prometheus-md-only",
"start-work",
"atlas"
"atlas",
"smart-failover"
]
}
},
Expand All @@ -98,7 +112,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -224,7 +248,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -350,7 +384,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -476,7 +520,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -602,7 +656,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -728,7 +792,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -854,7 +928,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -980,7 +1064,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1106,7 +1200,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1232,7 +1336,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1358,7 +1472,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1484,7 +1608,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1610,7 +1744,17 @@
"type": "object",
"properties": {
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down Expand Up @@ -1746,7 +1890,17 @@
"type": "string"
},
"model": {
"type": "string"
"anyOf": [
{
"type": "string"
},
{
"type": "array",
"items": {
"type": "string"
}
}
]
},
"variant": {
"type": "string"
Expand Down
59 changes: 59 additions & 0 deletions docs/features/smart-failover.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
# Smart Provider Failover

## 1. Overview
In multi-model environments, providers often hit **429 (Rate Limits)**, **Insufficient Balance**, or **Quota Exhaustion**.
The Smart Failover system provides an automated detection and recovery mechanism, ensuring uninterrupted service by switching to healthy fallback models instantly.

## 2. Key Features
- **Pipe Syntax (`|`)**: Minimalist fallback chain definitions.
- **Array Syntax (`string[]`)**: Equivalent to pipe syntax, easier to edit.
- **Instant Failover**: Aborts OpenCode's internal retry loops to trigger immediate model swapping.
- **Error Diagnosis (Best-Effort)**: Classifies common failures (rate-limit, quota, balance) via pattern matching.
- **Guardrails**:
- **Context Compatibility**: Skips fallbacks with insufficient context windows.
- **Probation Recovery**: After a cooldown elapses, a model becomes eligible again (PROBATION) and is cleared back to healthy after the session becomes idle.
- **Memory Safety**: Automatic cleanup upon session deletion.

## 3. Configuration
Smart Failover is enabled by default (unless you disable the `smart-failover` hook). Configure a fallback chain in `model` to use it.

### 3.1 Model Fallback Chain
You can define the fallback chain using either:

- **Pipe syntax** (string)
- **Array syntax** (string[])

Both forms are equivalent: the first entry is the primary model, and the rest are fallbacks.

### 3.2 Hook Toggle
If you need to disable it, add `smart-failover` to `disabled_hooks` in your `oh-my-opencode.json`.

### Example
```jsonc
{
"model": "openai/gpt-5.2-codex | google/gemini-3-pro"
}
```

### Array Example
```jsonc
{
"model": ["openai/gpt-5.2-codex", "google/gemini-3-pro"]
}
```

`model` can also be configured per-agent (e.g. `agents.Sisyphus.model`) and in category configs. Those locations also accept either pipe syntax or an array.

## 4. Default Behavior
- **Triggers**: Retry-loop detection (`session.status: retry`) and certain session errors (`session.error`) mark the current `provider/model` as unavailable and switch to the next available fallback.
- **Cooling + Backoff**: Retry-loop cooling uses a fixed 5-minute cooldown. Session-error cooling uses exponential backoff based on repeated failures.
- **Locking**: Balance/quota exhaustion signals lock a specific `provider/model` pair (model key) until reset.
- **Fallback Selection**: Only HEALTHY/PROBATION models are eligible; fallbacks with too-small context windows are skipped.

## 5. Limitations
- **Retry-After**: The implementation does not reliably receive response headers in events, so header-based cooldown is best-effort.
- **Probation**: Recovery is approximated by the session becoming idle, not a dedicated health-check request.

## 6. UI/UX
- **Notification**: A yellow toast appears: `⚠️ Switched to google/gemini-3-pro`.
- **Throttling**: Toasts are shown only once per session to prevent UI spam.
Loading
Loading