summaryrefslogtreecommitdiffstats
path: root/config.example.yaml
diff options
context:
space:
mode:
authordev <dev@localhost>2026-10-21 18:37:27 +0000
committerdev <dev@localhost>2026-10-21 18:37:27 +0000
commit9d40159cad3b55c753ed02e7a4dee8cc7b6d15eb (patch)
tree05572dd06f27e5817272f80e04bb7609f19b3f61 /config.example.yaml
parent3705a6141c42a828e714853d089c48e88a232de5 (diff)
downloadaichat-9d40159cad3b55c753ed02e7a4dee8cc7b6d15eb.tar.gz
Add request timeout and retry support to aichat HTTP clients
- Add timeout, retries, retry_delay_ms, and retry_max_delay_ms options to client extra config to bound long-running requests and retry failed attempts with exponential backoff. - Apply a total request timeout via reqwest .timeout(). - Implement retry with exponential backoff for non-streaming and streaming chat completions. - Disable reqwest-eventsource internal auto-reconnect so it does not loop forever; outer retry logic is now the single owner of retry behavior. - Document the new options in config.example.yaml.
Diffstat (limited to 'config.example.yaml')
-rw-r--r--config.example.yaml12
1 files changed, 12 insertions, 0 deletions
diff --git a/config.example.yaml b/config.example.yaml
index 4cac6c0..24e4cf9 100644
--- a/config.example.yaml
+++ b/config.example.yaml
@@ -116,6 +116,18 @@ clients:
# extra:
# proxy: socks5://127.0.0.1:1080 # Set proxy
# connect_timeout: 10 # Set timeout in seconds for connect to api
+ # timeout: 600 # Set total request timeout in seconds (0 or omitted disables)
+ # retries: 3 # Number of retries after the first failed attempt (0 disables)
+ # retry_delay_ms: 1000 # Base delay between retries in milliseconds (default 1000)
+ # retry_max_delay_ms: 60000 # Cap exponential backoff delay in milliseconds (default uncapped)
+
+ # A note on long-running providers (e.g. background/queued inference):
+ # A non-streaming request may hold one HTTP connection for minutes. If the
+ # connection is dropped by an intermediary, the client otherwise waits
+ # forever. Set `timeout` to bound each attempt, and `retries` to re-submit
+ # when an attempt fails. Be aware that retrying a non-idempotent request
+ # may resubmit the same work; prefer streaming (stream: true) for
+ # long-running requests so data flows continuously over the wire.
# See https://platform.openai.com/docs/quickstart
- type: openai