tag:blogger.com,1999:blog-14455177.post5589317281802694424..comments2024-03-07T18:57:25.977+01:00Comments on Mikael Ronstrom: NDB Checkpoints and research on In-Memory DatabasesMikael Ronstromhttp://www.blogger.com/profile/07134215866292829917noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-14455177.post-54597461539470919182020-10-06T19:38:58.746+02:002020-10-06T19:38:58.746+02:00We flush at least up to the latest epoch, but also...We flush at least up to the latest epoch, but also any pages that are ready<br />for writing. Even if a log record survives, it will only be restored if its<br />epoch is restored. So the restore will always restore a consistent checkpoint.Mikael Ronstromhttps://www.blogger.com/profile/07134215866292829917noreply@blogger.comtag:blogger.com,1999:blog-14455177.post-41245794471476816952020-07-17T10:59:32.592+02:002020-07-17T10:59:32.592+02:00Hi Mikael,
With this new checkpoint algorithm,the...Hi Mikael,<br /><br />With this new checkpoint algorithm,the log is still flushed to persistent storage every second and in a system-wide failure we can still recover to everything less the last second of data as before, right? If the answer is yes, I am wondering in the flush if you flush everything in the log available at that time or up to the boundary of an epoch. In other words, if there are committed transactions in the last epoch that hasn't completed, do you flush them too or just everything up to second to the last epoch?<br /><br />AlexAlex Ouhttps://www.blogger.com/profile/09780641053339854377noreply@blogger.com