关于VideoIndexer序列索引及多实例并行索引对速度影响的技术问询

Great questions about Video Indexer's processing behavior—let me break this down with practical context for your 1-minute video workflow:

Will Video Indexer complete indexing in submission order for videos sent every 1 minute? #

Short answer: No strict guarantee of first-in-first-out (FIFO) ordering, even with identical videos and consistent submission timing. Here's why:

• Resource allocation variability: Even if videos are the same length, minor differences in content (e.g., one has background noise, another has clear audio) can lead to different processing times. A video that needs more computational work (like complex speech-to-text for an accented language) might finish after a simpler one submitted later.
• Service queue logic: Video Indexer manages a pool of processing workers. Sometimes a later submission might get picked up by a free general-purpose worker before an earlier video that's waiting for a specialized worker (e.g., for face recognition).
• Transient errors & retries: If an earlier video hits a temporary glitch (like a network blip during processing) and triggers an automatic retry, it could end up finishing after a later video that processed smoothly.

That said, if all your videos are exactly identical in content and indexing settings, you'll likely see them complete in close to submission order—but don't rely on this for critical workflow sequencing. If strict ordering is a must, build a custom queue on your end that only submits the next video once the previous one's indexing status is marked as complete.

Does running multiple indexing instances affect indexing speed? #

Yes, but the impact depends on your setup and subscription limits:

• Within a single account/tier: Each Video Indexer tier has predefined concurrent processing limits. If you exceed this limit, extra jobs will just sit in a queue—they won't process faster. For example, if your tier allows 5 concurrent jobs, submitting 10 videos at once means 5 start immediately, and the rest begin as slots free up. Individual job speeds won't slow down, but overall throughput is capped by your tier.
• Across multiple accounts/regions: If you scale across separate Azure accounts or regions, each has its own resource pool. This lets you parallelize processing more effectively, boosting overall throughput since you're using independent capacity.
• Per-job overhead: Each indexing job has minimal setup/teardown time, which is negligible for 1-minute videos. Running multiple instances won't slow down individual jobs unless you hit extreme service-level resource contention (which is rare if you stay within your tier's limits).

内容的提问来源于stack exchange，提问作者naman Gupta