Elysia Tools
Navigation
Workflow Playbook
Define cache test cases, compare real response headers before and after a change, review TTL budgets and deliver redacted evidence with approval and rollback conditions.
Hubs
This workflow is for developers and operators approving one HTTP cache-policy change, not a broad performance audit. Keep the old configuration and an executable rollback plan outside this page. Record exact URL, method, query ordering, representation, time and client location. The query analyzer inventories parameters; it does not establish the CDN cache key. Changing a query to force a miss creates a different case, not a comparable repeat.
Classify public and personalized responses before choosing any shared-cache policy. Use only authorized synthetic accounts for isolation tests. Preserve private originals securely and prepare separate redacted copies before using tools or sharing evidence.
The header analyzer makes a server-side request and follows redirects. Record its final URL and method; it does not provide a complete per-hop trace or arbitrary authenticated browser state. The browser request tester runs from the browser, where CORS, exposed headers and restricted request headers limit observations. Its current fetch setup has no explicit cross-origin credentials inclusion option. Missing visible headers are not proof they were absent on the wire.
Use an authorized external capture for cases these clients cannot reproduce. Preserve exact request conditions and response content evidence. A server observation and a browser observation may differ by location and request context; keep them separate.
For a public test fixture, analyze public, max-age=60, s-maxage=600, stale-while-revalidate=30. This gives the analyzer a browser freshness input of 60 seconds, shared freshness of 600 and a 30-second revalidation allowance. These are an example policy, not universal recommended limits. Agree acceptable values with the content owner.
The analyzer accepts one header value and optional budgets. Its CDN notes are static and do not inspect rules, cache keys or purge state. Its stale total is an arithmetic review aid, not a guaranteed extra availability period or remaining freshness for an object with Age. Resolve duplicate directives and conflicts before approval. Do not turn a zero freshness calculation into a claim that nothing is stored; validation requirements and storage prohibitions are distinct.
After an operator deploys, repeat the same cases and compare raw headers, observed content, redirects and available cache diagnostics. Keep request times and client context. A 200 or 304 status alone does not prove an edge hit; timing alone does not prove correctness. Review relevant variation and authorized account isolation rather than extrapolating from one public response. Use external provider diagnostics when needed and mark untested cases pending.
Use text diff as a line-level comparison aid, retaining raw originals. If a cURL request helps reproduction, sanitize it before converting to HAR. The offline converter does not issue traffic or capture a response; label its result request-only. Never publish tokens or session cookies merely because the output is JSON.
Keep approved policy, configuration versions, test matrix, timestamps, observations, comparisons and reviewer decision together. Specify rollback triggers such as unintended cross-user content, excessive stale exposure or unmatched critical cases; a failed isolation check should block approval. These tools neither deploy nor purge nor execute rollback. Restoring headers alone may not remove already stored objects; the operator must define and verify the recovery procedure.
Save redacted shareable files locally and calculate SHA-256 after redaction; hashes identify copies, not correctness. Keep restricted originals separately. Use the related API-debugging workflow for request construction, network triage for transport incidents and technical SEO for URL publication checks. This handoff ends at a bounded cache-change decision, not a global CDN certification.
Workflow playbook
Inventory query parameters and define controlled URL/representation cases. Preserve the exact requested form; do not silently drop or reorder parameters.
Collect actual response headers and status from the authorized endpoint; record redirects and client context. Use browser testing only within its CORS and credential limitations.
Analyze the exact Cache-Control value against agreed browser, edge and stale budgets. Read status semantics and conflicts without treating a generic report as provider configuration.
Repeat matched requests and compare headers and content evidence. Optionally convert a sanitized cURL request to HAR as a request description, not as a recorded response.
Classify each response before proposing shared caching. Review authenticated and anonymous cases separately; do not infer safety from the public homepage.
Use the budget analyzer for directive arithmetic, not proof of CDN settings or a cache hit. Record actual headers and configuration independently.
Browser tests have CORS and header visibility limits. Server-side header collection uses a different client and network location; do not compare them as identical requests.
Record unresolved cases and rollback triggers, save redacted evidence and calculate file hashes. Deployment, purge and rollback are operator actions outside these tools.