[chatgpt] Add configurable request timeout by abrenk · Pull Request #20408 · openhab/openhab-addons

abrenk · 2026-03-20T12:15:33Z

The previously hardcoded 10 second timeout is now configurable as a thing-level parameter (requestTimeout) with an optional per-channel override. A channel timeout of 0 falls back to the thing default. This allows longer timeouts for local LLMs (e.g. ollama) or reasoning models that require more processing time.

An example would be a rule that generates a weather report based on sensor data.

Fixes #20203 (although already closed, but without any changes).

A cherry-pick to the 5.1.x branch would be nice.

lsiepel

Thanks for providing this patch.
Besides these comments, could youalso add the parameter to the readme.md ?

lsiepel · 2026-03-20T12:33:14Z

....chatgpt/src/main/java/org/openhab/binding/chatgpt/internal/ChatGPTChannelConfiguration.java


    public int maxTokens = 500;
+
+    public int requestTimeout = 0;


i would not expect this on the channelconfiguration as the PR describes this is on a per thing base.

From the PR description:

with an optional per-channel override.

The thing specifies the default value for all channels, but a channel can have its own.

Ah, yes i see now. I would opt to make this a @Nullable and not default to 0

...binding.chatgpt/src/main/java/org/openhab/binding/chatgpt/internal/ChatGPTConfiguration.java

lsiepel · 2026-03-20T12:35:29Z

...enhab.binding.chatgpt/src/main/java/org/openhab/binding/chatgpt/internal/ChatGPTHandler.java

        scheduler.execute(() -> {
            try {
-                Request request = httpClient.newRequest(modelUrl).timeout(REQUEST_TIMEOUT_MS, TimeUnit.MILLISECONDS)
+                Request request = httpClient.newRequest(modelUrl).timeout(DEFAULT_REQUEST_TIMEOUT_S, TimeUnit.SECONDS)


This does not respect the timeout set to the thing configuration, but uses the hard coded default.

This is the one-time request for /v1/models which should complete in milliseconds. If you want, I can of course change this, but the thing-level timeout is primarily for the HLI completions.

Yes i understand, With the provided value it is fine, but it mixes the default for a user configuration parameter and a setting that can be seen as more infrastructural related. Anyway, choose what you want, i have no strong opinion on this, just want to make sure you know.

bundles/org.openhab.binding.chatgpt/src/main/resources/OH-INF/i18n/chatgpt_de.properties

bundles/org.openhab.binding.chatgpt/src/main/resources/OH-INF/i18n/chatgpt_fr.properties

lsiepel · 2026-03-20T12:37:16Z

bundles/org.openhab.binding.chatgpt/src/main/resources/OH-INF/i18n/chatgpt.properties

file needs to be regenerated after other comments have been addressed.

lsiepel · 2026-03-20T12:39:10Z

...enhab.binding.chatgpt/src/main/java/org/openhab/binding/chatgpt/internal/ChatGPTHandler.java

-        Request request = httpClient.newRequest(apiUrl).method(HttpMethod.POST)
-                .timeout(REQUEST_TIMEOUT_MS, TimeUnit.MILLISECONDS).header("Content-Type", "application/json")
-                .header("Authorization", "Bearer " + apiKey).content(new StringContentProvider(queryJson));
+        return sendPrompt(queryJson, config != null ? config.requestTimeout : DEFAULT_REQUEST_TIMEOUT_S);


In the case requesttimeout is 0, this will gives issues

thing-types.xml specifies min="1".

Yes, but that does not enforce the value to be min="1" as we have file based / unmanaged configurations that can have any value, so a sanity check is needed.

What's the pattern then? Throw an exception or use the default value and log a warning?

Usually a binding verifies in the initialize method if all configuration parameters are valid. If not, the thing state is set accordingly.

...enhab.binding.chatgpt/src/main/java/org/openhab/binding/chatgpt/internal/ChatGPTHandler.java

lsiepel · 2026-03-20T12:40:11Z

bundles/org.openhab.binding.chatgpt/src/main/resources/OH-INF/thing/thing-types.xml

+				<label>Request Timeout</label>
+				<description>Timeout in seconds for this channel. Overrides the thing-level timeout. Set to 0 to use the thing
+					default.</description>
+				<default>0</default>


Suggested change

<default>0</default>

<default>10</default>

It should match the default set in the configuration.

The channel default is "0", i.e. to use the thing-level timeout value.

here i would remove the default tag to not set any value and leave it null

lsiepel · 2026-03-20T12:41:21Z

bundles/org.openhab.binding.chatgpt/src/main/resources/OH-INF/thing/thing-types.xml

 				<default>1</default>
 				<advanced>true</advanced>
 			</parameter>
+			<parameter name="requestTimeout" type="integer" min="0">


A zero time-out makes no sense, or in many cases it is used as a signal to not set any timeout.

Suggested change

<parameter name="requestTimeout" type="integer" min="0">

<parameter name="requestTimeout" type="integer" min="1">

That's exactly the case, as detailed in the <description>:

Set to 0 to use the thing default.

Not setting a timeout leaves it to the library, here it is set to the default time out. This is not a common pattern in openHAB and i would opt to reconsider this behavior.

This is a consequence of using "0" as the value to mean "use config from thing", the http library is always called with an explicit timeout. But as discussed above I will change this to be nullable and then the minimum value can be 1.

The previously hardcoded 10 second timeout is now configurable as a thing-level parameter (requestTimeout) with an optional per-channel override. A channel timeout of 0 falls back to the thing default. This allows longer timeouts for local LLMs (e.g. ollama) or reasoning models that require more processing time. Signed-off-by: Andreas Brenk <mail@andreasbrenk.com>

abrenk · 2026-03-20T13:34:23Z

I've removed the translations and added the requestTimeout to README.md.

abrenk added the enhancement An enhancement or new feature for an existing add-on label Mar 20, 2026

abrenk requested a review from kaikreuzer as a code owner March 20, 2026 12:15

lsiepel requested changes Mar 20, 2026

View reviewed changes

abrenk force-pushed the chatgpt-timeout branch from 309a52e to e682f4f Compare March 20, 2026 13:27

abrenk force-pushed the chatgpt-timeout branch from e682f4f to 6613c0a Compare March 20, 2026 13:33

	<parameter name="requestTimeout" type="integer" min="0">
	<parameter name="requestTimeout" type="integer" min="1">

Uh oh!

Conversation

abrenk commented Mar 20, 2026

Uh oh!

lsiepel left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abrenk Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lsiepel Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lsiepel Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abrenk commented Mar 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

abrenk Mar 20, 2026 •

edited

Loading

lsiepel Mar 20, 2026 •

edited

Loading

lsiepel Mar 20, 2026 •

edited

Loading