Ivan, I'm still not sure it is a good idea to upgrade metastorage automatically. Because we can't detect the correct charset the metastorage was created with, and at the same time we can't be sure the current charset is the correct one.
So, is there any guarantee the metastorage is consistent even if it was "upgraded" successfully? As I see, we just copy metastorage keys to a temporary one in key-by-key manner... and then do write-back to the original one. Seems, if smth goes wrong, the user may get both (original and temporary) stores broken. On Mon, Dec 20, 2021 at 5:27 PM Ivan Daschinsky <ivanda...@gmail.com> wrote: > Andrey, I believe that we already have all machinery to do migration safe. > See for > example > org.apache.ignite.internal.processors.cache.persistence.metastorage.MetaStorage#init > and > org.apache.ignite.internal.processors.cache.persistence.metastorage.MetaStorage.TmpStorage. > This machinery was introduced for slightly different task, but we can reuse > this for the current purpose. > > пн, 20 дек. 2021 г. в 11:53, Mikhail Petrov <pmgheap....@gmail.com>: > > > Thank you all for your replies! > > I got the idea and agreed with it. Based on the results of the > > discussion, I have filed a ticket [1]. > > I will try to investigate it. > > > > [1] - https://issues.apache.org/jira/browse/IGNITE-16157 > > > > On 16.12.2021 20:11, Ivan Daschinsky wrote: > > > Andrey, agree with you, good point. > > > > > > чт, 16 дек. 2021 г., 16:27 Andrey Mashenkov < > andrey.mashen...@gmail.com > > >: > > > > > >> Guys, > > >> > > >> I like the idea with a flag, but for a different purpose. > > >> I think it is easy to detect the issue (using the flag) when > > >> metastorage was created on a new version with a fixed charset, or on > an > > >> older version with the user-defined default. > > >> Regarding the flag, we can choose a new strategy forcing UTF-8, or > > fallback > > >> to the old one with defaultCharset and print a warning and > > recommendation > > >> in log. > > >> > > >> Adding any compatibility stuff is absolutely error-prone because if > you > > >> fail in the middle of restoring process, you will get broken > metastorage > > >> with keys in different charsets. > > >> At this point, there is no way to detect broken keys anymore. > > >> > > > > > -- > Sincerely yours, Ivan Daschinskiy > -- Best regards, Andrey V. Mashenkov
