lsemprini
commented
3 years ago - added a badly-needed Table of Contents (and see newly added script doc/make_toc.pl to automatically regenerate TOC any time)
- "Update Modes of HashDBM:" added more info about UPDATE_APPENDING
- "Multi-record Transaction:" clarified the text about rollback and ACID and added a link to the "broken" section
- "Restoring Broken Databases:" clearly define the threats, how tzkrw addresses them, how to get guarantees
- fixed a bunch of small spelling errors THINGS TO CHECK
- in "Restoring Broken Databases" under "HashDBM and TreeDBM" where I added a suggestion to set TuningParameters.align_pow to the disk block size when using UPDATE_APPENDING. That is because if TuningParameters.align_pow is 0, the end of the existing region might be on the same disk block as the start of the region for new records, and cause a problem with partial/corrupted writes on a power failure. And also, writing the last bucket value may be on the same disk block as the oldest existing records in the existing region. See what you think.
- In "Restoring Broken Databases" under "HashDBM and TreeDBM"I suggest to call the Synchronize() method with hard==true; I am not sure what the benefit of hard==false would be since I guess it would not cover most types of OS or hardware crashes. But we could talk about both if you think it is useful.