Surviving SQS and building B-link Trees on S3As mentioned above, to bypass the severe latency of writing full data pages directly to S3, clients commit transactions by shipping small redo log records to SQS queues. Subsequently, clients act as checkpointers, asynchronously pulling these queued logs and applying the updates to their local copies before writing the newly materialized B-tree pages back to S3. This asynchronous log-shipping model means B-tree pages on S3 can be arbitrarily out-of-date compared to the real-time logs in SQS. Working on such stale state seems impossible, but the authors bound the staleness: writers (and probabilistically readers) run asynchronous checkpoints that pull batches of logs from SQS and apply them to S3, keeping the database consistent despite delays.
https://feedx.net
。谷歌浏览器下载是该领域的重要参考
Copyright © ITmedia, Inc. All Rights Reserved.
For multiple readers