![]() ![]() ![]() Overall, I think the talk about splitting misses the point. The step-by-step strategy was supposed to take care of this organically, I guess. There's a bunch of random /proc entires too. In addition, the whole thing is driven primarily through ioctls on /dev/vzctl. Naturally, APIs to OpenVZ are largely ad-hoc. Split that into little pieces! I did not see any credible alternative from LXC (although I may be ignorant). Also, beancounters again: they hook into nearly every struct and method across the kernel, and they are essential for operations of VEs. That sounds great except for one thing: the way it's going, it will take longer than Linux's lifetime. The party line is to chip on OpenVZ piece by piece until everything is re-implemented as its LXC equivalent, at which time we can retire OpenVZ. Like the ptys, NFS, shmem, quotas (and a big chunk of networking, including AF_UNIX). Does it let me use OpenVZ (or LXC) to host my blog? Of course not! There's still 90% left. OK, so we adopted the virtual this and virtual that into the mainline. You know how the OS research in 1990s completely missed that microkernels are useless without the bulk of the kernel and ended toying with IPC into the irrelevance? Well, it looks to me that those who talk about splitting OpenVZ up and virtualizing piece by piece miss the same thing. The beancounter straddles the two.įirst, the size. Ignoring the checkpointing, the two biggest problems that I see is that of size and the APIs. It has some cruft, like unused and forgotten functions, but it was kept up to date (e.g. I only read it sporadically, so a dedicated engineer would probably consume it in about 5 to 7 days. My old post was right on the money: it took me a full month to read the OpenVZ diff (I finished by the beginning of November). Aurora is concerned about the cargo cult programming too: Please don't go add a load of fsync()s to your applications "just UPDATE: Pavel made my point better than I did. UPDATE: LWN article "Better than POSIX?" has a title ending with a question mark - an old journalistic trick! I realize he only talks about files that application deems permanent, but in reality it's cancer that's going to poison every application through the cargo cult programming practices and will stay with us forever (necessitating ruining fsync for everyone, which is exactly his plan, apparently). The only acceptable criterium is the reasonable use. Hey, manpage says we can! This doesn't make the argument any more relevant. Doing just that is the favourite trick of morons who insist on returning an error from close(2), too. Next step: fcntl(F_FULLFSYNC).Īlso, hiding behind the letter of a manpage is plain wrong. Įspecially fabulous is how, in the same breath, he proposes disabling fsync() in kernel. I want to use Finance::QuoteHist::Google to retrieve the google finance data.Requesting all applications to add fsync() before close() is madness. Yahoo format for indices looks like this: ^DJUSAE and the database must save in this format.google expects DJUSAE. in this case it must used a string of defined symbols like backpopulate above. I need someone to add two new functions in beancounter that tell the script to alternatively source the data from google instead of yahoo default when invoked with a google command line switch and save to stockprices tableīeancounter backpopulate -prevdate '1 year ago' -date 'yesterday' ^DJUSAE ^DJUSHB. Recently Yahoo stopped providing updates for Dow Jones Industry indices listed below. I am using a modified version of Finance::Beancounter (attached) that retrieves stock information from Yahoo.
0 Comments
Leave a Reply. |