nick
d0022402ae
mirror: Start our timeout watcher from the first, not second, transfer
2013-08-14 15:29:24 +01:00
nick
28fff91af1
flexnbd: add a mirror-speed command to change mirror->max_bytes_per_second
...
It's not actually honoured yet, and ideally, you'd also be able to set it as
part of the initial setup: "flexnbd mirror ... -m 4G". remote_argv for the
mirror case would need to become x=y z=w format first, though.
2013-08-14 13:33:02 +01:00
nick
385c9027db
flexnbd status: display mirror->max_bytes_per_second as mirror_speed_limit
2013-08-14 13:30:25 +01:00
nick
b73081e417
One more fix
2013-08-13 16:22:44 +01:00
nick
cc468b0b17
control/mirror: Use uint64_t and strtoull to get max_Bps into the mirror
2013-08-13 12:30:18 +01:00
nick
7128fcc901
control: Output abandoned mirror state
2013-08-13 12:29:53 +01:00
nick
45355666f7
mirror: And another abandon fix
2013-08-12 16:14:53 +01:00
nick
8a294e5ee0
mirror: fix abandon
2013-08-12 15:54:49 +01:00
nick
c6764b0de1
mirror: abandon signals are now honoured outside of the remote end being readable / writable
2013-08-12 15:30:21 +01:00
nick
41facd2ccf
Branch merge
2013-08-09 17:07:06 +01:00
nick
f6456349f7
Backed out changeset e58ff57b5e2d
...
Slows tests down
2013-08-09 17:06:56 +01:00
nick
9f4fbe782c
Branch merge
2013-08-09 17:03:25 +01:00
nick
8c750a5e9d
listen: Allow longer gaps between transfers
2013-08-09 17:02:58 +01:00
nick
64702d992d
Minor fixes here and there
2013-08-09 17:02:33 +01:00
nick
c2df38c9d3
mirror: Use libev to provide an event loop inside the mirror thread
...
We're doing this so we can implement bandwidth controls sanely.
2013-08-09 17:02:10 +01:00
nick
754949d43f
bitset: Add a bitset_run_count_ex that lets you learn the value of the bits in the run
2013-08-09 16:49:38 +01:00
lupine
1a966ca0be
bitset: Prove that bitset operations with len=0 don't underflow
2013-07-26 17:09:21 +01:00
nick
f590f8ed3c
status: Add migration_speed ( bytes per second ) and migration_duration( seconds ) to the migration output
2013-07-26 11:50:01 +01:00
nick
bc9ce93648
bitset: squash one more bug
2013-07-25 10:58:50 +01:00
nick
a5870b8e9b
Remove a stray debugging statement
2013-07-25 10:14:14 +01:00
nick
bed8959d47
bitset: Fix large runs
2013-07-24 17:42:08 +01:00
nick
5c59a412af
flexnbd-proxy: ensure upstream cooldown is applied when read init from upstream fails
2013-07-24 16:01:38 +01:00
nick
253cee5a10
flexnbd: Acknowledge new return type of bitset_run_count
2013-07-24 15:08:29 +01:00
nick
7de22a385e
flexnbd: clients should be MADV_RANDOM, rather than MADV_SEQUENTIAL
2013-07-24 14:18:23 +01:00
nick
14db3315ca
non-debug builds get -O2 for impressive bitset speedups
2013-07-24 12:34:36 +01:00
nick
efe9eaef7c
bitset: A more-efficient bit(set)_run_count
2013-07-24 12:03:24 +01:00
nick
f8fd4e0437
bitset: Actually enable an optimization in bit_set/clear_range
...
Previously, we were setting bits up to the first byte boundary,
memset()ing to the last byte boundary, then ignoring the memset()
and resetting every single bit up to the last one individually,
from where the first for-loop left off.
This should be *at least* nine times faster.
2013-07-24 11:19:52 +01:00
nick
9a37951aaa
bitset: Use uint64_t everywhere to avoid possible integer overflows
...
Hasn't been a problem in practice, mind.
2013-07-24 10:34:22 +01:00
nick
d18423c153
tests: Fix a couple of compile warnings
2013-07-23 17:22:23 +01:00
nick
1b0fe24529
test: Add some tests for bitset_run_count
2013-07-23 17:13:40 +01:00
nick
5c5636b053
flexnbd mirror: If the final run would be longer than the file size, truncate to file size
...
This fixes migrations of images that are not exactly divisible by 4096
2013-07-23 11:00:51 +01:00
nick
afe76debf7
flexnbd status: Actually output pass statistics
2013-07-08 14:27:04 +01:00
nick
f4bfc70a4b
flexnbd status: Add current pass clean/dirty byte statistics
2013-07-08 13:51:15 +01:00
nick
b29ef6d4de
flexnbd status: Avoid a possible NULL dereference reading migration status
...
While the mirror mutex is taken, the mirroring can be abandoned and serve->mirror
set to NULL, so we need to lock around reading information from serve->mirror
2013-07-08 13:32:14 +01:00
nick
dee0bb27d6
flexnbd status: Add the size of the backing file, in bytes
...
This will be handy information if you're querying flexnbd for migration
stats, particularly.
2013-07-08 10:11:18 +01:00
nick
f556f298b1
flexnbd status: Add current migration pass to the status output if we're migrating
2013-07-08 09:58:31 +01:00
nick
55b452ebef
Fix tests for new killswitch argument
2013-07-03 10:04:08 +01:00
nick
9f34752842
flexnbd: Make the killswitch runtime-selectable
...
We're not actually using it in production right now because it doesn't
shut its sockets down cleanly enough. This is a better option than
reverting the functionality or keeping production downgraded until
we sort out a handler that cleanly closes the sockets.
2013-07-03 09:56:35 +01:00
nick
81d41f567d
proxy: Reduce the reconnect cooldown from 15 seconds to 3.
...
Exponential backoff would be better, but that's OK
2013-06-20 10:26:34 +01:00
nick
89fd18f6f0
proxy: Add a 30-second timeout for requests in-flight to upstream
...
It's a little more complicated than that, actually. For the various
states that involve reading from, or writing to, the upstream fd,
if the amount of time spent in that state is > 30 seconds, we reconnect
to the server and resend the request.
we also introduce a 15-second reconnect dampener to keep us from stressing
things unduly. This may need to be decreased, or turned into an exponential
backoff, at some point.
2013-06-19 16:36:19 +01:00
nick
3c56ba0af6
proxy: Fix a comment
2013-06-19 11:27:09 +01:00
nick
2a9884e9e9
proxy: Fix the prefetch code
2013-06-19 11:18:52 +01:00
nick
1afea5c73d
proxy: Respect the REQUEST_MASK
2013-06-19 11:18:22 +01:00
nick
62bdad2a6e
ioutil: Add a bit more debug output to iobuf_read/write
2013-06-19 11:17:46 +01:00
nick
cd0a1f905f
proxy: The minor optimisation bugs if needle is not advanced on iobuf_read()
2013-06-19 11:16:35 +01:00
nick
2156d06368
proxy: DRY up some code
2013-06-18 16:58:39 +01:00
nick
b14bba36ec
proxy: Set proxy->upstream_fd before calling proxy_finish_connect_to_upstream
...
The only thing this affects is a log message
2013-06-18 15:58:38 +01:00
nick
f5c434f21c
proxy: Initial move to event-loop proxy model.
...
Building with -DPREFETCH is currently broken, I'm sure, but otherwise
this version seems to be feature-complete compared to the previous one,
albeit wordier. Upcoming: cleanups
2013-06-18 15:37:39 +01:00
nick
662b9c2d07
readwrite: Expose a couple of points of functionality
2013-06-18 15:36:15 +01:00
nick
197c1131bf
tests: Tell us which offset fails
2013-06-18 15:35:24 +01:00
