Proxy Session Affinity: One Request or a Whole Workflow?
Choose whether proxy affinity belongs to one request, one authenticated workflow, or a longer-lived account before enabling automatic rotation.
A login, a redirect, and a report download are three requests. For the user, they are one job. Rotating the exit between every request can make that job harder to reason about even when each proxy works.
Start by defining the unit that owns the route. A request-level picker is suitable for independent operations. For those public operations, the free proxy list offers candidates when public intermediaries are permitted; it supplies no sticky-session contract. A workflow-level choice keeps related steps on an intentional path until the workflow finishes or a defined recovery decision replaces it.
Cookies and proxy affinity are different state
An application cookie belongs to the website's session rules. A provider's sticky-session identifier, when offered, belongs to its proxy service and documented connection or account policy. They do not become one setting merely because both use the word session.
An ordinary CONNECT intermediary cannot read the website's encrypted Cookie header and use it as a routing key. If a service promises cookie-driven behavior, find out which visible interface carries that cookie and whether TLS terminates there. Do not assume the website's private cookie is automatically available to a tunneling gateway.
Keep a small workflow record: chosen endpoint, provider session identifier if used, application session identifier, and the condition that ends the affinity. Avoid putting account secrets into routine record keys.
Plan the uncomfortable case
If the selected route fails halfway through, decide whether the workflow can safely continue on another exit, needs to re-establish its application session, or must stop. A failed response does not always mean a submitted operation was never processed.
Neither permanent stickiness nor constant rotation is universally correct. Choose the scope from the application's consistency requirements, then apply request pacing and service limits to the whole workflow. More exits do not remove those requirements.