node-fetch-retry

Node Module for performing retries for HTTP requests.

It is a wrapper around node-fetch library. It has default retry logic built in as described below, as well as configurable parameters. It also has built-in support for Apache OpenWhisk actions, adjusting the timeout to reflect the action timeout.

Installation

1npm install @adobe/node-fetch-retry

Usage

This library works the same as the normal fetch api, but with some added features.

Default Behavior

Without configuring any parameters, the retry behavior will be as follows:

• retry for 60s
• retry inital delay of 100ms with exponential backoff, configurable as a multiplier defaulting to 2
• retry only on 5xx response
• retry on all FetchError system errors
  • see node-fetch error handling: https://github.com/node-fetch/node-fetch/blob/main/docs/ERROR-HANDLING.md
• socket timeout of 30s

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url);
5}

This example uses only custom headers and will use default retry settings:

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        headers: {
6            'custom-header': '<<put custom header value here>>'
7        }
8    });
9}

Optional Custom Parameters

All the retry options are configurable and can be set in retryOptions in the options object passed to fetch.

Parameter	Format	Description	Environment variable	Default Value
retryMaxDuration	Number	time in milliseconds to retry until throwing an error	NODE_FETCH_RETRY_MAX_RETRY	60000 ms
retryInitialDelay	Number	time in milliseconds to wait between retries	NODE_FETCH_RETRY_INITIAL_WAIT	100 ms
retryBackoff	Number	backoff factor for wait time between retries	NODE_FETCH_RETRY_BACKOFF	2.0
retryOnHttpResponse	Function	a function determining whether to retry given the HTTP response. Can be asynchronous	none	retry on all 5xx errors
retryOnHttpError	Function	a function determining whether to retry given the HTTP error exception thrown. Can be asynchronous	none	retry on all FetchError's of type system
socketTimeout	Number	time until socket timeout in milliseconds. Note: if socketTimeout is >= retryMaxDuration, it will automatically adjust the socket timeout to be exactly half of the retryMaxDuration. To disable this feature, see forceSocketTimeout below	NODE_FETCH_RETRY_SOCKET_TIMEOUT	30000 ms
forceSocketTimeout	Boolean	If true, socket timeout will be forced to use socketTimeout property declared regardless of the retryMaxDuration. Note: this feature was designed to help with unit testing and is not intended to be used in practice	NODE_FETCH_RETRY_FORCE_TIMEOUT	false

Note: the environment variables override the default values if the corresponding parameter is not set. These are designed to help with unit testing. Passed in parameters will still override the environment variables

Custom Parameter Examples

This example decreases the retryMaxDuration and makes the retry delay a static 500ms. This will do no more than 4 retries.

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: {
6            retryMaxDuration: 2000,  // 30s retry max duration
7            retryInitialDelay: 500,
8            retryBackoff: 1.0 // no backoff
9        }
10    });
11}

This example shows how to configure retries on specific HTTP responses:

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: {
6            retryOnHttpResponse: function (response) {
7                if ( (response.status >= 500) || response.status >= 400) { // retry on all 5xx and all 4xx errors
8                    return true;
9                }
10            }
11        }
12    });
13}

This example uses custom socketTimeout values:

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: {
6            retryMaxDuration: 300000, // 5min retry duration
7            socketTimeout: 60000, //  60s socket timeout
8        }
9    });
10}

This example uses custom socketTimeout values and custom headers:

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: {
6            retryMaxDuration: 300000, // 5min retry duration
7            socketTimeout: 60000, //  60s socket timeout
8        },
9        headers: {
10            'custom-header': '<<put custom header value here>>'
11        }
12    });
13}

This example shows how to retry on all HTTP errors thrown as an exception:

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: {
6            retryOnHttpError: function (error) {
7                return true;
8            }
9        }
10    });
11}

Disable Retry

You can disable all retry behavior by setting retryOptions to false.

1const fetch = require('@adobe/node-fetch-retry');
2
3async main() {
4    const response = await fetch(url, {
5        retryOptions: false
6    });
7}

Disabling retry behavior will not prevent the usage of other options set on the options object.

Additional notes on retry duration

If the fetch is unsuccessful, the retry logic determines how long it will wait before the next attempt. If the time remaining will exceed the total time allowed by retryMaxDuration then another attempt will not be made. There are examples of how this works in the testing code.

Apache OpenWhisk Action Support

If you are running this in the context of an OpenWhisk action, it will take into account the action timeout deadline when setting the retryMaxDuration. It uses the __OW_ACTION_DEADLINE environment variable to determine if there is an action running.

Behavior: If retryMaxDuration is greater than the time till the action will timeout, it will adjust the retryMaxDuration to be equal to the time till action timeout.

Contributing

Contributions are welcomed! Read the Contributing Guide for more information.

Licensing

This project is licensed under the Apache V2 License. See LICENSE for more information.