Reef boards in lab have model:reef but are really electro or basking |
|||||
Issue descriptionIt looks like the testing for reef (electro/basking) is effectively disabled in Goldeneye, I just turned on testing for these models in GE, but I am afraid they will never actually test if we don't have models labeled correctly in the lab. For example if we look at http://cautotest/afe/#tab_id=view_host&object_id=7670 it shows it is board:reef model:reef today, but it shows phase:PVT_ELECTRO and variant:electro, so I think it is really an electro? If we look for platforms in cautotest called electro we see only one http://cautotest/afe/#tab_id=view_host&object_id=2145 However something is still appearing to kick off bvt runs on these DUTs https://viceroy.corp.google.com/chromeos/dut_utilization?board=reef&model=&pool=managed%3Abvt&status=Running&is_locked=False&topstreams=35&build_config=kevin-paladin&build_number=&duration=8d&experimental=False&mdb_role=chrome-infra&milestone_version=&refresh=-1&scheduler_host=cros-full-0036&sentinel_host=chromeos-server156&staging_master=chromeos-staging-master2&waterfall=chromeos Should we label the systems with variant:electro as model:electro and similarly with basking? We seem to have more electro DUTs than basking so maybe we ultimately only test on electro in this case...
,
Jun 20 2018
From what we understand we were suppose to deploy only electro NOT basking https://b.corp.google.com/u/0/issues/64805465#comment3
,
Jun 20 2018
It looks like somebody re-jiggered the reef builder to be a unibuilder with models named "reef", "basking", and "electro". Previously, when unibuilds were less stable, we had problems with the labels on basking and electro. I don't remember the details, but I think we ended up manually relabeling to make sure tests could run. Now that the build has changed, I suspect we just need to manually re-label again. We might want to revisit how the "model" label automatic detection is working. I think there's logic that prevents the "model" label from changing. That was needed at one time, but we may have reached the point where that's no longer the right thing to do.
,
Jun 21 2018
,
Jun 28 2018
Any update on this? Whom is the right person to decide if we should change the labels, and to actually change the labels?
,
Jun 28 2018
Based on the way that reef was setup originally electro and basking should be used interchangeably. We have both boards deployed to cover the single deployment for reef. If this needs to be changed please let me know soon, as we will most likely need to purchase a large amount of devices.
,
Jun 28 2018
They are almost all electro, how short are we on electro? I guess the real root problem here is that Goldeneye does not show results for reef currently, so we have to check other resources like Stainless. We don't have reef as a model of reef in the build views (seems like a bug). https://screenshot.googleplex.com/pFqpakcGy45 https://screenshot.googleplex.com/8uXtkmtKhDn We also don't seem to be able to use the lab board value to redirect results from 'reef' boards into the rows for electro and basking. If we really want to keep the electro and basking as only reef from a lab perspective we can fix this in other ways I guess.
,
Jul 26
Following up from bug 867624.
,
Jul 26
Cleaned the reef labels using:
"
#!/bin/bash
# Replace all instances of a label in the lab with a different label.
# USE WITH CAUTION!
#Set to the old label to replace.
OLD_LABEL="model:reef"
#Set to the new label we want.
NEW_LABEL="model:electro"
#Find all DUTs with a particular label
DUTS=$(atest host list -b ${OLD_LABEL} | cut -d ' ' -f1)
#Remove extra Host entry
DUTS=$(echo ${DUTS} | cut -d ' ' -f2-)
echo "Found the following DUTs to change labels on:"
echo ${DUTS}
for DUT in ${DUTS}; do
echo "Replacing ${OLD_LABEL} with ${NEW_LABEL} on ${DUT}"
atest label add ${NEW_LABEL} -m ${DUT}
atest label remove ${OLD_LABEL} -m ${DUT}
done
echo "Done..."
"
Confirmed results in cautotest/.
|
|||||
►
Sign in to add a comment |
|||||
Comment 1 by jkop@chromium.org
, Jun 20 2018Status: Assigned (was: Untriaged)