1 |
On Sat, Aug 22, 2020 at 04:15:38AM +0000, Caveman Al Toraboran wrote: |
2 |
> ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ |
3 |
> On Saturday, August 22, 2020 12:10 AM, Grant Taylor <gtaylor@×××××××××××××××××××××.net> wrote: |
4 |
> |
5 |
> > There is some nebulous area around what that actually means. But the |
6 |
> > idea is that the receiving server believes, in good faith, that it has |
7 |
> > committed the message to persistent storage. Usually this involves |
8 |
> > writing the message to disk, probably via a buffered channel, and then |
9 |
> > issued system calls to ask the OS to flush the buffer to disk. |
10 |
> |
11 |
> just to double check i got you right. due to |
12 |
> flushing the buffer to disk, this would mean that |
13 |
> mail's throughput is limited by disk i/o? |
14 |
> |
15 |
> or did i misunderstand? |
16 |
> |
17 |
> i sort of feel it may suffice to only save to |
18 |
> disk, and close fd. then let the kernel choose |
19 |
> when to actually store it in disk. |
20 |
|
21 |
When an M.T.A. encounters mail, the content of the mail will first exist in the |
22 |
M.T.A.'s local memory, in a buffer. Before sending an "OK" to the sending |
23 |
server, it should first make an attempt to write it to disk, through an fwrite |
24 |
(stdio) or write (POSIX) call. At that point, it is, in theory, the kernel's |
25 |
choice if and when it is _actually_ written to disk, but if one of the |
26 |
aforementioned functions return a success code, the M.T.A. has done its bit, and |
27 |
can consider the message "safely stored". |
28 |
|
29 |
-- |
30 |
|
31 |
Ashley Dixon |
32 |
suugaku.co.uk |
33 |
|
34 |
2A9A 4117 |
35 |
DA96 D18A |
36 |
8A7B B0D2 |
37 |
A30E BF25 |
38 |
F290 A8AA |