1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
|
TODO list (in order of priority)
---------
- more charsets for file, symlink and directory names
- complete set of operations
- handling of most file system metadata
- ownership: how is this done depending on
per-user or root mounts? think about security
here!
- extended attributes
- pooling of PQconn
- thread-synchronization (we can ommit the -s flag)
- use of asynchonous read/writes
- what is a transaction? couple to fsync/fdatasync
for files? other ideas? customizable?
- how to store binary data?
- bytea complete: needs memory on client and server,
ok for small files
- split bytea (in blocks, as mysqlfs): should give
good average performance
- BLOBs: streamable, no referencial integrity though,
have a weak security model
=> most likely a split bytea with tunable blocksize
(with a small block cache on the side of pgfuse
is the way to go)
- is there a way to implement streams (append to
file, or sequential read) efficient together
with random access? Use COPY TO/COPY FROM?
=> most likely not worth the hazzle. Most streams
(but for BLOBS) are mere abstractions in the
database binding of the programming language
=> sticking to traditional libpq C way
- use prepared statements, measure performance gain
- establish self-containment (with respect to
a temporarily unavailable Postgresql server)
- minimal SELinux support, i.e. one fix security context
per mount point
unique: 8, opcode: GETXATTR (22), nodeid: 7, insize: 68
unique: 8, error: -38 (Function not implemented), outsize: 16
lsetxattr("file.sgml", "security.selinux", "system_u:object_r:httpd_sys_content_t:s0", 41, 0) = -1 EOPNOTSUPP (Operation not supported)
- investigate SELinux races
- make some performance tests, see http://archive09.linux.com/feature/127055
|
