AZ-204<\/a> connect to AI-200. Both involve APIs, events, authentication and data interactions, but the AI backend often adds expensive model calls and retrieval jobs whose runtimes and error patterns are less predictable than ordinary CRUD operations.<\/p>\nDesign retry policy around the type of failure<\/h2>\n
Not every failure deserves an immediate retry. A transient network error may clear quickly. A service returning a rate-limit response may require exponential backoff and respect for the service’s recommended delay. A malformed document may never succeed without input correction. An unauthorized request should not be retried indefinitely as though another attempt will grant permissions. Classifying failures protects downstream services and prevents expensive repeated model inference.<\/p>\n
Retry budgets should operate across the workflow, not only within one worker. If a function calls a client library that already retries, and the message infrastructure retries the entire function again, one customer request may produce many more downstream attempts than expected. Track retry attempts and processing state with the job identifier. If a partial update occurred, determine whether it is safe to resume from a checkpoint or whether a compensating action is required.<\/p>\n
A practical document workflow stores authoritative job state separately from the message broker. The queue coordinates work; the database records which stage is complete and the result users may retrieve. If a worker disappears, the next worker can inspect state and continue or safely reprocess. This separation is especially important where the model response is expensive, generated output needs approval, or the application must prove exactly which source version was processed.<\/p>\n
Scale on workload pressure without losing cost control<\/h2>\n
A queue growing from ten to ten thousand documents suggests that processing capacity and arrival rate have diverged. Adding workers may reduce backlog, but only if downstream services can absorb additional traffic. Model APIs may enforce request or token limits; databases have capacity limits; large documents consume memory. A scaling rule based solely on queue length can amplify overload by launching more simultaneous expensive operations. Throughput planning should consider processing time, concurrency, allowed retry rate and downstream quotas.<\/p>\n
The cost model changes, too. If a failed document triggers repeated inference calls, the business may pay for unsuccessful work. If a worker runs continuously with almost no requests, its hosting plan may not match the traffic pattern. A correct design places limits around document size, token budgets, processing duration and user-visible wait time. Metrics should distinguish requests accepted, processing started, succeeded, retried, dead-lettered and explicitly rejected for a known reason.<\/p>\n
Security travels with the work item. A message containing only a file ID can still allow unauthorized processing if the worker does not verify tenant ownership. Producers and consumers need appropriate Azure identities and permissions, and sensitive text should not be copied into general operational logs. Queue encryption or network controls do not replace application-level authorization on the source document.<\/p>\n
Trace one job through the entire system<\/h2>\n
Imagine that an uploaded financial report causes a function to emit a processing command. A worker retrieves the approved file, splits it into chunks, generates embeddings, writes retrieval records and emits a completion signal. The user later asks for a summary. If the summary is incomplete, operators need to determine whether the upload succeeded, whether the queue received the command, whether the worker ran, whether an embedding call failed, and whether completion was published prematurely. A correlation ID shared across the stages makes that diagnosis achievable.<\/p>\n
Construct one failure drill: deliberately send a malformed file, exhaust a downstream quota or make the database temporarily unavailable. Measure how many attempts occur and when the job becomes visibly failed or recoverable. Inspect the dead-letter queue and verify that replaying the corrected work does not create duplicate output. AI-200’s messaging material is fundamentally about making these workflows dependable and understandable when real systems fail, not about memorizing three overlapping Azure service descriptions.<\/p>\n","protected":false},"excerpt":{"rendered":"
A user uploads a hundred-page document and expects an AI application to summarize it. The work might take longer than a normal HTTP request should remain open, and it may involve document conversion, embedding generation, database updates and a final notification. Treating all of that as one synchronous call makes failures difficult to recover from […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10,1],"tags":[],"class_list":["post-2940","post","type-post","status-publish","format-standard","hentry","category-microsoft","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/posts\/2940","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/comments?post=2940"}],"version-history":[{"count":1,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/posts\/2940\/revisions"}],"predecessor-version":[{"id":3443,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/posts\/2940\/revisions\/3443"}],"wp:attachment":[{"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/media?parent=2940"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/categories?post=2940"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.exam-topics.net\/blog\/wp-json\/wp\/v2\/tags?post=2940"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}