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
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
|
0.9.7 (1232780)
- bug fix for -MD and -MF
- bug fix for gcc 4.6 plugins
0.9.6 (1158638)
- fix installation issues
- add a value type of ticks in /proc - and make it long long
- fix kFreeBSD build
- fix run-failure with glibc 2.11
- allow tgz as extension
- support more local args (bnc#625621)
0.9.5 (1083362)
- fix: close the file descriptor (novell bug 525799)
- log PID even if no prefix is specified
- allow get_msg() to fetch data from the kernel even if timeout is 0.
- clean up event loop of gcc invocation
- Install an "icerun" symlink that will serialize commands through
icecream's local daemon
0.9.4 (961598):
- fix compilation warnings
- don't leak file descriptor to create-env
- don't use the shell to call simple commands
- assert current_kids is > 0 before decrementing (it's unsigned)
- also check for EAGAIN (needed on OS X)
- add ICECC_LOGFILE to get trace() output into a file
- Fix compilation when using -Wformat -Werror=format-security flags
- make a special exception for - in rest args (bnc#495786)
- Fix way icecream changes permissions of /var/cache/icecream
0.9.3 (926474):
- Fix FTBFS with GCC 4.4
- Explicitly set group ownership and permissions on the base cache
dir and target dir in case root has a umask of e.g. 0077.
Otherwise nobody user won't be ableto untar.
- make sure that a too strict umask doesn't ruin the game
fixes https://bugzilla.novell.com/show_bug.cgi?id=446329
- some create-env fixes for fedora
0.9.2 (879112):
- On OS X, append the release number to Darwin (Darwin8 for Tiger, Darwin9 for Leopard)
as their icecream environments are not compatible
- Add accessor for the hostname of the scheduler, so that we can show it in a tooltip in
icemon in the future.
- Only localize job if we find -Wa,.*-a[a-z]*= in an argument
- also add accessor for network name
- if called as "icecc" (i.e., not through a symlink) and the first argument starts with a slash,
use that as the command to run.
0.9.1 (822102):
- trying to support gentoo x86_64
- -mcpu=native has to compile locally (gentoo bug 183586#c13)
- don't compile C files with C++ compile if ICECC_CXX is set
0.9.0 (807014):
- create a working env on OS X
- don't hardcode tar path
- implement load informations on OS X
- make it work again on non-Linux platforms: Jobs still reported their
environment platform as i386 etc, even if they came from Darwin or
BSD (thus the scheduler never finds a matching compile server, or
wrongly directs a non-Linux client to use a Linux compile server)
- fixing the init script
- patch by Tais M. Hansen to ignore directories
- compile on OS X 10.5
- some preparations for rsync support
- fix valgrind warning
- add configure check for librsync
- don't kick daemons that run as user, just don't give them
any jobs. this helps the people that just don't trust the
daemon to be able to at least send out jobs
- remove the now pointless "runasuser" option. it will just
not accept any jobs then
- add log message when the local daemon hangs again
- accept TERM=dumb as well
- implement support for scheduling local non-compile jobs.
For example: you can add a symlink named "meinproc4" pointing
to icecc to your path, and it will automatically schedule
the job and then invoke meinproc4 on it locally.
the advantage is that you can use cmake -j50 without having
to worry that there are going to be 50 meinproc jobs killing
your machine.
- do an asynchronous connection to the scheduler:
* helps when the scheduler is down, as the daemon
is otherwise locked up then and can't schedule
local jobs
* helps when the scheduler is flaky and blocked
itself on network (like in the suse network).
- fix connect if scheduler is known
- send jobs to a monitor non-blocking
- implement host blocking
- switch monitor connections to bulk transfer
- fix double deletion when control connection disconnects
- add greeting banner that gives a rough statistic (for pretty
monitoring graphs :)
- protocol cleanup
- fix error handling
- adding more API for better monitor behaviour
- various protections against malicious requests
- add a try-restart target for the rpm package
- always read the scheduler message in complete before trying
to fetch a message. fixes a daemon hangup (surprising noone noticed
this before)
- various code cleanups
- don't flood the scheduler
- implement bulk sending
- reduce overhead by not sending a full path
- implement much shorter monitor messages
- reenable cork handling on linux only. helps a lot
- fix job statistics in monitor
- check if the compile server sent us a message before die'ing another
death
- avoid killing the daemon on protocol errors
- support "quit" on the normal daemon port as well
- ext/hash_set is not needed and does not exist anymore
- DoS protection for the scheduler
- avoid that daemons hang around in environment installation
state forever
- fix daemons being overrun by wrongly assigned jobs
while they're busy installing
- add a segfault catcher
- don't allow tiny environments - they are very likely broken
- always relogin to the scheduler even if environment
transfer failed
- implement proper handling for PCH files for e.g. Qt compilation
- fix compilation with glibc 2.8
0.8.0 (657903):
- Find md5 on FreeBSD
- apply patch to support cygwin a bit better
- write a pid file. patch by Ismail Doenmez
- precompiled header support, needed for compiling Qt 4.3
- add support for --no-remote. patch by Matt Gruenke
- Make it work on DragonFlyBSD.
- provide SuSEfirewall2 information
- Make the result of make dist-gzip usable.
- update node name each time we login to the scheduler.
- treat -combine and -fdump* as always local
- Merged make-it-cool branch:
* implement job preloading
* asynchronous install of new environments
* uses tcp keepalive for network connection tracking
- make it work on OS X
* don' hardcode tar path
* create environments correctly on OS X
* report load correctly on mach kernels (Darwin)
* prepend os to the machine type if it's not linux
(resulting in e.g. Darwin_PowerMacintosh)
0.7.14 (583722):
- fix current_kids getting out of sync if send_scheduler
failes in an unfortunate moment
- fix reporting the error
- add more to the internals dump
- trying to make the scheduler kicks a little bit less frequent
- if the compiler is supposed to keep temp files, then we
have to do it locally
0.7.13 (583347):
- use file -L to follow symlinks in create-env
- fix an easy valgrind error
- make handling of the return values consistent
- if the client is in WAITCOMPILE then it's waiting for some
other host to compile, not this one. So job == 0 -> crash
- fix crash if the scheduler goes away while daemon transfers
environment
- apparently sometimes gcc can hang forever in some
cases. By the time the client disconects, we know
we don't have to wait any longer, because there is
nobody left caring about the result.
- update node name each time we login to the scheduler.
Fixes tons of "linux" hosts appearing in the icecream
monitor
0.7.12 (r581454):
- fixing error handling when scheduler restarts
- do not waitpid before we're sure we read all of g++'s
output otherwise g++ waits for us to read and we're
waiting for g++ to finish -> deadlock
0.7.11 (r581076):
- fix a crash in the daemon when the scheduler was gone while
local jobs were waiting for finishing
- separate stat handling from ping handling
to avoid excessive stat/ping loops
(increases internal version number)
- only reset scheduler ping time if the scheduler ping'ed us
- even when we can't determine native environment, we
can still use the daemon for inter-process locking
instead of falling back to file locking.
- quicker reap of dead daemons
- improved load guessing
- fix stupid logic bug in ping tracking
0.7.10 (r580794):
- handle errors in installing environments correctly
- block daemons that have full discs
- add -pipe to the command line to reduce disk usage
- fix cancelling of jobs awaiting a remote job (were hanging
in scheduler forever)
- if ICECC=no is set, don't try to be clever
- adding two more flags for local compilation (profile
feedback related)
- flush debug files before every fork to avoid dups in
log output
- be stricter in what messages are required to keep the
daemon<->scheduler communication intact (network drops
again)
0.7.9 (r580498):
- scheduler will ping the daemon periodically,
daemon will disconnect if not pinged from time
to time (to avoid network drops being unnoticed)
- removed some debug output
- make logging more consistent
- increased internal protocol version (still compatible)
- try even harder to avoid races in daemon
- rework how clients are catched
- some random cleanup
- remove the "this should be an exception" output to
avoid failing configure checks for nothing
- make sure the assembler is dead before we cleanup object
files
0.7.8 (r579795):
- fd leak fix, which caused a deadlock under
certain conditions
- rework some FreeBSD patches that caused problems
- fix race between select and SIGCHILD
0.7.7 (r574260):
- clear all internal maps
- patches by Frerich to support FreeBSD
- avoid busy loop in bizarre conditions
- found another case of endless hanging jobs
- some process fixes
0.7.6 (r561764):
- report all daemon error message to the client
- fix support for biarch machines that have only one
environment installed with another environment with
the same name and different architecture being around
- work-around a bug in older libstdc++
- ensure nobody uses -u root
0.7.5 (r561480):
- don't block while looking for a scheduler over broadcast search.
- fix logrotate
- immediately kick broken daemon connections
0.7.4 (r559927):
- Add -fpreprocessed argument
- Fix daemons getting stuck after a while with 100% CPU
- fix for kubuntu where libs are deeply nested
such as /lib/tls/i686/cmov/libc.so.6
- find the compiler even though icecc wasn't in the path
- fix deletion of old environments
- a lot more error handling
- fix assertion failures in daemon upon sigchild delivery
- use -fPIC also for C files
- avoid race between local and remote daemon about who has
the right exit status
- rename the package to icecc
0.7.3 (r552930):
- more work on error handling
- corrected icecc --help output
- handle local icecream load correctly
- reduce the number of load updates from daemon to scheduler
- early launch the compiler even before the whole preprocessed
file has been received
- handle low-latency networks better
- builtin colorgcc
- let the daemon schedule local load even without scheduler (important
for multi processor machines)
0.7.2 (r549574):
- split out libicecream-devel for the monitor
- introducing ICECC_PREFERRED_HOST (=mybigfathost) to let the scheduler
ignore all rules - this is meant to be a debugging aid
- fix network performance between client and local daemon
- replaced create-env with icecc --build-native (respecting the user's
$PATH by default)
0.7.1 (r548846):
- Monitor is an extra package now
- don't make non-existant cross compilers fatal, but just use
the other environments in ICECC_VERSION
- always handle aborted syscalls
- update minilzo
- never exit() or return from fork()'ed code paths,
0.7.0 (r547196):
- WARNING: the protocol is not compatible with earlier versions, old clients
will refuse to connect, so update the full farm
- protocol reworked to no longer require a client->scheduler connection
(the daemon will proxy it all)
- all local jobs are now scheduled by the local daemon to make sure
multiprocessor machines are handling as many jobs as they have jobs
configured (# processors). They used to be "scheduled" by lock file
- fixed grave bug that removed all daemon environments when the cache
grew over limit
- new telnet interface command: internals to output daemon states
- stricter error handling to avoid suprises
- SIGHUP to daemon and scheduler will reopen the log file to allow logrotate
- a restart of the daemon and scheduler won't flush the log file, but append
to it
- create-env will create .tar.gz now - the client allows environments to be
.tar, .tar.bz2 and .tar.gz
|