Fuego 1.0 wiki

Fuego release 1.0.9-notes

Contents:

Here is a list of problems encountered during the 1.0.9 release testing:

/fuego-core/engine/scripts/ftc-unit-test.sh
- can't download clitest
- uses /userdata instead of /fuego-ro for accessing board stuff
had to create testplan_fuego (for fuego-specific tests)

Jenkins jobs have the wrong TESTNAME
- e.g. docker.testplan_fuego.Functional.fuego_board_check has "TESTNAME=fuego_board_check" instead of 'fuego_test')
  - fuego-create-jobs doesn't read test.yaml file
- see if fuego-create-jobs uses the test name for the base script name
(worked around) can't find test with default plan in spec list
- code is missing my patches for ignoring 'testplan_default'
- you get an error from ovgen.py
docker board is missing reboot, according to Functional.fuego_board_check
- scan_form_items.sh reports HAS_REBOOT=0
  - but docker has /sbin/reboot
- test is run as user 'jenkins' which doesn't have /sbin in the path
- Functional.fuego_board_check is correct!

(fixed) parser.py gets an exception
- it's still using FUEGO_PARSER_PATH, instead of FUEGO_CORE environment var.
plot.png is empty
- missing metrics.json file
- multiplot mapping requires a specific metric name patterns (leading item of same name, to chart the data in a single chart?
  - plotting is VERY fragile!!
    - no one can debug this stuff
    - matplotlib producing empty graphs is hard to debug
FIXTHIS - should make plot API much easier, or at least document the rules better

item ^	priority ^	notes ^
add nod and job scripts to ftc	high
automatically create a view for each board	medium	use a command line option on fuego-create-node (--make_view)
eliminate DISTRIB	medium	automatically detect, and use board vars
ensure logs are linked even for failed tests	high	if log exists, there should be link, failure or not
fix ftc device dictionary	defer	used for board check and pre-populating options
handle set -e differently	medium	<empty cell>
handle timeouts automatically	medium	should be in test and board, not in plan - but depends on spec, transport, build time etc
have test do description-setting	medium
move /fuego-ro/conf/boards to /fuego-ro/boards	medium
ov_transport_connect/ov_transport_disconnect	medium	Sounds like Jan-Simon will do it
put post_test process kill in test_cleanup	high	need optional test_cleanup, that calls kill_procs, that calls ov_rootfs_kill
rename all scripts to fuego_test.sh	low
support ftc run-test	low	BUILD_TIMESTAMP won't work with separate post_test, but we're not doing that
support phase reordering	defer
support separate build phase	defer	don't want to hold board reserved during build, for lava integration and handling timeouts better
testing by Daniel	high	not available until next week
testing by Jan-Simon	high	not available until next week
use spec name instead of plan name in job name	medium	Use <board>.<spec>.<testname>
web file transport	defer	could be useful for some interesting target cases

List of functions that have post_test arguments:

Benchmark.netpipe: kills iperf (should be NPtcp)
- netpipe test doesn't use benchmark.sh!! (there is no call to post_test)
Functional.LTP.Open_Posix: kills run-posix-option-group-tst.sh run-tests.sh run-test
- this test is not in the 'next' branch
Functional.netperf: kills netperf
- netperf doesn't use functional.sh!! (there is no call to post_test)
Functional.LTP.Filesystem: kills run-test
- this test is not in the 'next' branch
(OK) Benchmark.fio: fills fio
Benchmark.OpenSSL: kills openssl
- Openssl doesn't use benchmark.sh (there is no call to post_test)
(OK) Benchmark.lmbench2: kills lmbench lat_mem_rd par_mem
(OK) Benchmark.IOzone: kills iozone
(OK) Benchmark.iperf: kills iperf
Benchmark.Java: kills java
- put in test_cleanup, but didn't call kill_procs
- there could be other java programs running - kill string is too generic
(OK) Benchmark.Interbench: kills interbench
(OK) Benchmark.gtkperf: kills gtkperf