Age | Commit message (Collapse) | Author |
|
Broke up the tests based on the top-level Java packages (see
openembedded/mauve/all-packages.txt).
Running all the Mauve tests as one single test means that if a timeout
occurs we get zero results. By having individual tests we can have much
shorter timeouts and, should a test timeout, then only that test will
have no results as opposed to Mauve in its entirety. This change also
makes it practical to schedule the tests onto more devices using
multi-node testing.
Change-Id: I310282af072f40223f0998a1f5fda79b481dd584
Signed-off-by: Andrew McDermott <andrew.mcdermott@linaro.org>
|
|
|
|
This forces the VM to use the template interpreter - which is what we
want to initially validate when running with OpenJDK 8.
Signed-off-by: Andrew McDermott <andrew.mcdermott@linaro.org>
|
|
Signed-off-by: Andrew McDermott <andrew.mcdermott@linaro.org>
|
|
The current version of the test harness has too many issues with:
spawning many threads (5000+)
timeout doesn't work properly
timeout leaves the pipe process in a hung state
VM size grows to 13+ GB
eventual stack overflow when restarting after a timeout
and as such the tests do not run to completion on Open Embedded
aarch64 using the ARM fastmodel.
Rather than trying to fix these issues in the existing Harness I have
duplicated the file (now HarnessLinaro.java) where all of the issues
listed above now seem to be fixed. However, given that this hasn't
had much soak testing (1 day) and I really want to get the LAVA tests
to run to completion, we will need to revisit these changes to see if
they are suitable for pushing upstream.
This commit now invokes the tests using HarnessLinaro[.java].
Signed-off-by: Andrew McDermott <andrew.mcdermott@linaro.org>
|
|
Note: a lot of the tests are commented out until the core Java classes
pass without issue.
Signed-off-by: Andrew McDermott <andrew.mcdermott@linaro.org>
|