-
Notifications
You must be signed in to change notification settings - Fork 20
/
RELEASE.txt
264 lines (232 loc) · 11.2 KB
/
RELEASE.txt
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
# This number should track progressive changes in this software, and correspond to git tags
# denoting releases. It is entirely independent of the the Pavilion's VERSION.
RELEASE=2.4
# Release History
## 2.5 Pre-release notes
- In the 'modules' section of 'run' or 'build' in test configs, modules specified without a version
no longer verify that the default was loaded.
- Added the ability to configure module wrappers in test/host configs.
- This breaks backwards compatibility with existing module wrapper plugins. The 'load()' method
now takes an addtional, required 'name' argument.
- Added the ability to extend lists via inheritance across all test configs.
- Added `pav series sets` and `pav series cancel` commands.
- `pav series history` is now `pav series `states`.
- Added `-c` 'overrides' to `pav series run`, and an overrides section to the series config.
- `pav cancel` no longer cancels series, just tests.
- Series now only generate tests as needed, rather than all in advance, and handles test creation,
build, and run errors gracefully.
- The `SCHED_RUNNING` state is now `SCHED_STARTUP` - existing scheduler plugins that use
`SCHED_RUNNING` should still work.
- Added an '--all-passed' option to the results command
- Added `pav log states` command, which lists all states a test has had.
## 2.4 Release Notes
- Minimum supported python version is now 3.6
- All IO or Processor intensive parts of Pavilion have been made multi-threaded or multi-process,
vastly speeding up many Pavilion operations.
- Tests now have id's of the form '<label>.<id>'. The default label is 'main',
so tests from your 'main' (PAV_CONFIG_DIR defined) pavilion config directory
can leave off the label. Labels aren't saved, but determined when configs
are loaded.
- Every config directory can (and should) now have a 'config.yaml' that sets the label for tests
built from that config directory, as well as the working_directory for tests created
from those configs. If this does not exist, a default label is applied to the tests,
and the general pavilion.yaml defined working directory is used.
- Pavilion has removed all permissions management, except to operate with 'shared_group' as
the default group. Given multiple working directories, it is now up to the user
to set sticky bits on those directories if specific permissions are desired for those tests.
- Test series files now have a 'repeat' config item, to specify the number of times the tests
in the series should be repeated.
- The code for test Series has been refactored and greatly cleaned up.
- Scheduler plugins have been completely revamped (see below).
- Machines can be 'chunked', and those chunks can be permuted over for smaller run sizes that
still cover the entire system.
- Chunk node distributions can be contiguous, random, distributed, and more.
- You can specify a percentage of the system for chunk size or node count.
- All builtin plugins are now directly integrated into the Pavilion code, to speed plugin search
time.
- Test nodes are (generally) chosen before a test is created, so you can permute over
`sched.test_node_list` and similar.
- Added `pav series ls command`
- Added `pav config` commands.
- Backwards compatibility breakso
- Scheduler configuration has been unified under the 'schedule' key in test configs (see below).
- Imports for the base plugin libraries have changed (but those libraries have not).
- Tests can no longer have skip conditions that are deferred.
- The '--wait' argument of the run command has been removed.
- Tests can no longer set test specific group and permissions
- The 'restart' test series config item is no longer supported.
- Many 'sched' variables in Pavilion test configs have changed names. See
`pav show sched --vars slurm` (or 'raw').
- The Slurm MPI scheduler plugin has been merged with Slurm. Use the slurm specific 'mpi_cmd'
parameter to use `mpirun` instead of `srun`. Additional `mpirun` options are also available.
- In test series files, the 'series' key is now 'test_sets'.
- The `pav series` command is now 'pav series run`
### Upgrade guide
#### Scheduler config
This update will unfortunately break most tests and plugins, but the fixes are simple.
- The individual scheduler config sections are now just 'schedule'.
- Most schedule parameters are universal, and most kept their names from when they
were under the `slurm` section.
- The big exception is `num_nodes` is now just 'nodes'.
- It no longer supports node ranges (Use 'min_nodes')
- Non-universal arguments are still in scheduler specific sections, but under `schedule`.
See `pav show sched --config` for the full format.
In general, this means you should rename all 'slurm' sections in test/host/mode configs to
'schedule', and all 'num_nodes' to just 'nodes'. Some config values (like 'up_states') will need
to be moved as well. Also `nodes` no longer takes a range. Use `min_nodes` instead.
For example, given the following:
```
mytest:
run:
cmds: "echo 'I am an example!'"
slurm:
num_nodes: 50
partition: foo
up_states: ['IDLE', 'NOT_ON_FIRE']
```
would become:
```
mytest:
run:
cmds: "echo 'I am an example!'"
schedule:
nodes: 50
partition: foo
slurm:
up_states: ['IDLE', 'NOT_ON_FIRE']
```
#### Scheduler Variables
Scheduler variables have also changed considerable.
- Almost all are universal.
- All slurm `alloc_*` variables are now `test_*`.
- Most can be converted directly.
## 2.3 Release Notes
- Added 'flatten_results' option to the base Pavilion config. Allows for
producing a separate result log line for each 'per_file' result, making
charting in Splunk significantly easier. The results can be significantly
larger, however, expecially on large machines.
- Added unified test run filtering in 'status' and 'result'.
- Added the 'list' command for searching test runs and series
- Added 'maint' command for maintanence operations.
- Results now include a 'uuid' key - A completely unique id for each test.
- Added spack options to test config. Allows for spack packages to be built
and/or loaded inside of a pavilion test. Requires a valid spack path to
be provided in the Pavilion config under the new key `spack_path`.
- Modules may now have nested paths (True lmod module organization)
- Updated permissions management.
- Added test series files for organizing multiple test runs.
- Added 'pav graph' command.
- Added 'pav show hosts' and 'pav show modes' commands.
- Switched to using epoch time for all timestamps.
- Test run attributes are now indexed for faster 'pav status' and 'pav result'
searches.
- Backwards compatibility breaks
- 'fn' and 'n' result fields have been merged into 'per_file'.
- The 'status', and 'result' commands now search all available test runs
and select recent (1 day) tests by the current user on the current
machine. For the old behavior, use ``pav status last``, etc.
- Old tests will lack some expected properties for searching,
as that information has been consolidated into the 'attributes' file
which they won't have.
- 'test_node_list' in the results has been replaced with the much shorter
'test_node_list_short' key.
- The results command '-l' (--show-logs) option is now '-L'.
- The results command '--json' argument and '--oneline' argments were removed.
Use --summary instead.
## 2.2 Release notes
- All new test config parser.
- Variable references are now 'expressions'.
- In addition to variables, they can contain math and functions.
- `'hello {{world * 2}}'`
- Assuming 'world' has a value of 5, would be: 'hello 10'
- Iterations (formerly sub-strings) can now contain more than one variable.
This results in all combinations of the variables being used.
- `'Greetings: [~hello {{name}}{{num}}~,]'`
- With `name: ['bob', 'janet']` and `num: [1,2]` would result in:
- 'Greetings: hello bob1, hello bob2, hello janet1, hello janet2'
- per test permissions settings
```yaml
mytest:
group: secret_group
umask: '007'
```
- Backwards compatibility breaks
- All duplicate keys in configs are now detected and cause errors.
- Default values may no longer be specified using '{{foo?}}'
- Dashes are no longer allowed in Pavilion variable names.
- `'{{name-len}}'`
- You can no longer have nested 'sub-strings' (now called iterations).
- Not everything can be escaped in Pavilion strings anymore. This is now
limited to all normal python escapes
(https://docs.python.org/2.0/ref/strings.html),
plus '[', '{', and '~'.
- The results section is now 'result_parse'
- Result parsers plugins now have explicit argument default and validation settings,
and the _check_args method must return the modified argument dictionary.
(rather than None before)
```yaml
mytest:
variables:
unit_base: 100
result_parse:
regex:
speed:
regex: 'speed (\d+)'
result_evaluate:
speed_normalized: "speed / {{unit_base}}"
another_thing: '"some don''t string"'
```
- build.source_location has changed.
```yaml
mytest:
build:
source_path: mytest.zip
source_url: http://github.com/hpc/mytest/latest.zip
source_download: latest
- Build are now built in their final location, rather than in a tmp
location. This fixes the issues with symlinks (as they no longer
move, so they can be absolute), and issues with non-relocatable
builds.
- New config keys:
-
```yaml
mytest:
maintainer:
name: Paul Ferrell
email: pferrell@lanl.gov
```
## 2.1.2
- Fixed some regressions from the 2.1.1 release.
## 2.1.1
- Fixed speed of draw_table with new algorithm
- Fixed some name conflicts with the 'per_fullname' and 'per_name' result parser options.
- Fixed issue with pipe exceptions in pav results
- Fixed issue with slurm scheduler node selection.
- Fixed clean command exceptions
- Updated reserved result parser key list. Renamed 'errors' to 'pav_result_errors'.
- and many smaller bug fixes.
- Updated yaml_config version.
- Updated the ls and cat commands, merged the tree command into ls.
- Moved build code into its own module
- Added include/exclude_nodes option to slurm.
- Added a 'build' command. Like the 'run' command, but only builds.
- Added '--rebuild' option to both build and run (and the capability to do so)
- Added better build tracking output and verbosity under the 'run' command.
## 2.1
- Quite a few bugfixes.
- Simplified time output (and dropped pytz dependency)
- Updated regex parser.
- Migrated documentation to Readthedocs and restructured text format.
- Added API documentation
- Added some tutorials
- Add ls and cat commands
- Moved exception tracking to a single log file.
- Added the concept of hidden tests.
- Re-organized how deferred variables work.
- run.sh files are now regenerated when a test _run's
- test variables are now saved in the test run directory.
- Fixed RUN_COMPLETE files. They are now made when a test is finished,
fails, is cancelled, and in most error cases.
## 2.0
- All new Pavilion 2.0 redesign
- Pavilion was completely redesigned and reimplemented for this release.