[chromium-dev] Re: Flakiness. Please help.
On Fri, Aug 7, 2009 at 6:46 AM, Marc-Antoine Ruel mar...@chromium.orgwrote: On Thu, Aug 6, 2009 at 5:05 PM, Jeremy Orlowjor...@chromium.org wrote: On Thu, Aug 6, 2009 at 2:02 PM, Nicolas Sylvain nsylv...@chromium.org wrote: On Thu, Aug 6, 2009 at 1:05 PM, Scott Violet s...@chromium.org wrote: Not sure, perhaps Huan could answer that. That said, --enable-dcheck certainly works on the Chromium release builds from the buildbot: http://build.chromium.org/buildbot/continuous/LATEST/ . Yes, --enable-dcheck is supposed to be disabled in Google Chrome build. Disabled or optimized out (so all that code isn't in the resulting binary). Hopefully the latter. Optimized out. Actually, it only goes away on Windows at the moment. http://crbug.com/16512 Quick summary from the bug: logging.cc uses a CPP symbol of OFFICIAL_BUILD, which only exists in Windows builds (it's not in GYP it's in props files). mmoss can also expand on what it would take to clean up/use OFFICIAL_BUILD since I believe he's looked into that. Anyway, I made one run at getting it to go away (CL and revert cl are in the bug), but the Mac build started hanging in unittests when these got optimized away, I've made one pass looking for things in DCHECKs that have side effects without any luck. TVL --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
Maybe this is a stupid question, but where can I find test-expectations.txt? I looked in the tree and didn't see it. -Paul Wicks On Wed, Aug 5, 2009 at 11:12 PM, Peter Kastingpkast...@google.com wrote: On Wed, Aug 5, 2009 at 10:10 PM, Eric Seidel esei...@chromium.org wrote: Do we have a list of flakey tests? I feel like we used to have one... Every test in test-expectations.txt marked PASS FAIL or PASS FAIL CRASH or some similar set of multiple expectations is flaky, or at least thought to be so. PK --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
src/webkit/tools/data/layout_tests btw, run_layout_tests.* can be used to run stuff (and lint the exceptions file) On Wed, Aug 5, 2009 at 11:26 PM, Paul Wicks pwick...@gmail.com wrote: Maybe this is a stupid question, but where can I find test-expectations.txt? I looked in the tree and didn't see it. -Paul Wicks On Wed, Aug 5, 2009 at 11:12 PM, Peter Kastingpkast...@google.com wrote: On Wed, Aug 5, 2009 at 10:10 PM, Eric Seidel esei...@chromium.org wrote: Do we have a list of flakey tests? I feel like we used to have one... Every test in test-expectations.txt marked PASS FAIL or PASS FAIL CRASH or some similar set of multiple expectations is flaky, or at least thought to be so. PK --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
Oops...I failedit's actually src/webkit/tools/layout_tests Also the file uses an underscore, not a dash...it's test_expectations.txt J On Wed, Aug 5, 2009 at 11:28 PM, Jeremy Orlow jor...@chromium.org wrote: src/webkit/tools/data/layout_tests btw, run_layout_tests.* can be used to run stuff (and lint the exceptions file) On Wed, Aug 5, 2009 at 11:26 PM, Paul Wicks pwick...@gmail.com wrote: Maybe this is a stupid question, but where can I find test-expectations.txt? I looked in the tree and didn't see it. -Paul Wicks On Wed, Aug 5, 2009 at 11:12 PM, Peter Kastingpkast...@google.com wrote: On Wed, Aug 5, 2009 at 10:10 PM, Eric Seidel esei...@chromium.org wrote: Do we have a list of flakey tests? I feel like we used to have one... Every test in test-expectations.txt marked PASS FAIL or PASS FAIL CRASH or some similar set of multiple expectations is flaky, or at least thought to be so. PK --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
On Wed, Aug 5, 2009 at 9:44 PM, Peter Kastingpkast...@google.com wrote: * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) I'm willing to be a valgrind contact. - Dan --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
One easy suggestion in helping catch bugs is to run Chrome with --enable-dcheck . This'll prompt if you hit a DCHECK in release builds and hopefully help isolate crashes before the fact. -Scott On Wed, Aug 5, 2009 at 9:44 PM, Peter Kastingpkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these P1? Because until we prove they can't affect Chrome itself, they potentially can, and Chrome crashes are always P1. They affect stability and security both. (5) If you have commit rights, go ahead and TBR test-expectations changes you're confident of. I even suggest using --force if the tree is closed. Updating expectations is like fixing bustage, it helps the tree go green faster and thus is almost always desirable. If you don't have commit rights, send your review to the WebKit sheriff. *** Your reward for reading this far: * At the end of the quarter, I will nominate for a peer bonus every Googler who puts something meaningful about flakiness/test failures/the other stuff above on their OKRs, accomplishes it, and sends me a note pointing that out. * At the end of the quarter, I will nominate for commit access every non-Googler who sends me a pointer to ten
[chromium-dev] Re: Flakiness. Please help.
To clarify, doesn't --enable-dcheck only work on chromium release builds you built yourself and not official builds of Google Chrome? On Thu, Aug 6, 2009 at 10:15 AM, Scott Violet s...@chromium.org wrote: One easy suggestion in helping catch bugs is to run Chrome with --enable-dcheck . This'll prompt if you hit a DCHECK in release builds and hopefully help isolate crashes before the fact. -Scott On Wed, Aug 5, 2009 at 9:44 PM, Peter Kastingpkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these P1? Because until we prove they can't affect Chrome itself, they potentially can, and Chrome crashes are always P1. They affect stability and security both. (5) If you have commit rights, go ahead and TBR test-expectations changes you're confident of. I even suggest using --force if the tree is closed. Updating expectations is like fixing bustage, it helps the tree go green faster and thus is almost always desirable. If you don't have commit rights, send your review to the WebKit sheriff. *** Your reward for reading this far: * At the end of the quarter, I will
[chromium-dev] Re: Flakiness. Please help.
Not sure, perhaps Huan could answer that. That said, --enable-dcheck certainly works on the Chromium release builds from the buildbot: http://build.chromium.org/buildbot/continuous/LATEST/ . -Scott On Thu, Aug 6, 2009 at 12:59 PM, Antony Sargentasarg...@google.com wrote: To clarify, doesn't --enable-dcheck only work on chromium release builds you built yourself and not official builds of Google Chrome? On Thu, Aug 6, 2009 at 10:15 AM, Scott Violet s...@chromium.org wrote: One easy suggestion in helping catch bugs is to run Chrome with --enable-dcheck . This'll prompt if you hit a DCHECK in release builds and hopefully help isolate crashes before the fact. -Scott On Wed, Aug 5, 2009 at 9:44 PM, Peter Kastingpkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these P1? Because until we prove they can't affect Chrome itself, they potentially can, and Chrome crashes are always P1. They affect stability and security both. (5) If you have commit rights, go ahead and TBR test-expectations changes you're confident of. I even suggest
[chromium-dev] Re: Flakiness. Please help.
On Thu, Aug 6, 2009 at 1:05 PM, Scott Violet s...@chromium.org wrote: Not sure, perhaps Huan could answer that. That said, --enable-dcheck certainly works on the Chromium release builds from the buildbot: http://build.chromium.org/buildbot/continuous/LATEST/ . Yes, --enable-dcheck is supposed to be disabled in Google Chrome build. Nicolas -Scott On Thu, Aug 6, 2009 at 12:59 PM, Antony Sargentasarg...@google.com wrote: To clarify, doesn't --enable-dcheck only work on chromium release builds you built yourself and not official builds of Google Chrome? On Thu, Aug 6, 2009 at 10:15 AM, Scott Violet s...@chromium.org wrote: One easy suggestion in helping catch bugs is to run Chrome with --enable-dcheck . This'll prompt if you hit a DCHECK in release builds and hopefully help isolate crashes before the fact. -Scott On Wed, Aug 5, 2009 at 9:44 PM, Peter Kastingpkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these
[chromium-dev] Re: Flakiness. Please help.
Do we have a list of flakey tests? I feel like we used to have one... On Wed, Aug 5, 2009 at 9:44 PM, Peter Kasting pkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these P1? Because until we prove they can't affect Chrome itself, they potentially can, and Chrome crashes are always P1. They affect stability and security both. (5) If you have commit rights, go ahead and TBR test-expectations changes you're confident of. I even suggest using --force if the tree is closed. Updating expectations is like fixing bustage, it helps the tree go green faster and thus is almost always desirable. If you don't have commit rights, send your review to the WebKit sheriff. *** Your reward for reading this far: * At the end of the quarter, I will nominate for a peer bonus every Googler who puts something meaningful about flakiness/test failures/the other stuff above on their OKRs, accomplishes it, and sends me a note pointing that out. * At the end of the quarter, I will nominate for commit access every non-Googler who sends me a pointer to ten patches relating to the above items that they have posted for review, and who doesn't otherwise have
[chromium-dev] Re: Flakiness. Please help.
On Wed, Aug 5, 2009 at 10:10 PM, Eric Seidel esei...@chromium.org wrote: Do we have a list of flakey tests? I feel like we used to have one... Every test in test-expectations.txt marked PASS FAIL or PASS FAIL CRASH or some similar set of multiple expectations is flaky, or at least thought to be so. PK --~--~-~--~~~---~--~~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~--~~~~--~~--~--~---
[chromium-dev] Re: Flakiness. Please help.
Ojan is working on the tool for the layout tests. First bits are already checked in. :DG On Wed, Aug 5, 2009 at 10:10 PM, Eric Seidelesei...@chromium.org wrote: Do we have a list of flakey tests? I feel like we used to have one... On Wed, Aug 5, 2009 at 9:44 PM, Peter Kasting pkast...@google.com wrote: THIS MAIL APPLIES TO YOU Flakiness is growing. Smash it before it gets bigger, and keep it smashed. *** The MOST IMPORTANT section in this gigantic mail: PLEASE spend some of every workday (or each week at least, if you can't spare time each day) looking at test failures, flakiness, valgrind/purify/coverity bugs, crashes, and/or memory bugs. Make it a goal to get an average of one line in the test-expectations file removed each day. If you're a Googler, put it on your OKRs (now, not sometime tomorrow). * DON'T wait for someone to assign bugs to you or ask for your help * DON'T wait for a team fixit week (those haven't worked) * DON'T wait for someone else to solve the problems * DON'T wait until after your current project is finished * DON'T wait until you have worked on WebKit HELP, even if it's just a little, even if it's not your core competence. We currently have hundreds upon hundreds of failing or flaky tests. We can dramatically reduce this quickly but ONLY IF YOU HELP. This is an investment not only in the quality of Chrome but in the team's ability to move fast, so help here doesn't just improve the quality of Chrome, but also the derivative of the quality :) (If you do not know how to do anything above and need handholding, e-mail me and I will help you. It's OK to be ignorant.) *** Next, how you should help keep the tree green at all times: * If you ever look at the buildbot and see red, and there's no explanation in the build status, ask what's going on on #chromium. Ping the sheriffs specifically (they're listed in the upper-right corner). If you do not get an answer about ownership within a few minutes, close the tree (if you have the rights to) or ask someone to close it. THE TREE SHOULD NOT BE OPEN WITH RED THAT NO ONE OWNS. Help the sheriffs out with this -- they can't watch every second. Closed trees suck; unowned bustage sucks more. Be hard-nosed. * Yes, even purify, valgrind, and reliability bot redness. If you can't figure out what to do with these, try pinging erikkay for purify issues and huanr for reliability issues. (Not sure who a good general valgrind contact is.) * If you ever look at the buildbot and see orange (unexpected pass), especially in the WebKit LayoutTest bots, ping the WebKit sheriff (the calendar is linked from the top of http://dev.chromium.org/developers/how-tos/webkit-merge-1 ; I don't know whether it's world-readable). If he wasn't aware of it, agree between you on who will deal with it. Orange alone is not reason to close the tree, but it should NOT be ignored. * DON'T IGNORE TESTS BECAUSE THEY WENT GREEN ON THE NEXT CYCLE. If they're really fixed by someone's commit, that should be easy to determine. Otherwise, they're flaky, and we NEED to mark them as such, not just leave them. *** Finally, how to help if the LayoutTest bots are red or orange: (1) Try and determine if the test(s) are consistently passing/failing unexpectedly, or if they're flaky. Make sure you look at all the different bots to see which OSes are affected. (2) Update src/webkit/tools/layout_tests/test-expectations.txt. Look for the test(s) in question. Often, flaky tests will already be in there as failing or flaky for one OS, and need to have more added; or they will be marked flaky (FAIL PASS) and need CRASH added. If they're not there, add a line. (3) Ensure the test(s) have a bug on file. Note the bug on the expectation. (4) If any tests are crashing (flaky or not), they're high-priority and someone needs to triage them. Today, dglazkov was WebKit sheriff and was having me mark these bugs as P1, Mstone-3, owner:dglazkov. I'm not sure whether the Right Thing is to assign them to the WebKit sheriff or still to him (feel free to comment, dglazkov!). Why are these P1? Because until we prove they can't affect Chrome itself, they potentially can, and Chrome crashes are always P1. They affect stability and security both. (5) If you have commit rights, go ahead and TBR test-expectations changes you're confident of. I even suggest using --force if the tree is closed. Updating expectations is like fixing bustage, it helps the tree go green faster and thus is almost always desirable. If you don't have commit rights, send your review to the WebKit sheriff. *** Your reward for reading this far: * At the end of the quarter, I will nominate for a peer bonus every Googler who puts something meaningful about flakiness/test failures/the other stuff above on their OKRs, accomplishes it, and sends me a note pointing that out. * At the end of the quarter, I will nominate for commit access