I have seen a Selenium test suite (for a very AJAXy application) that had random unexplainable nondeterministic failures. Actions like "wait for element with id=foobar to appear" and then the immediate next action "click on element with id=foobar" would suddenly say it couldn't find the element any more---but only sometimes and not always.
People spent hours trying to find the root cause, without any luck.
With this type of error, particularly with an ajaxy application, you want to use 'wait for element to be clickable' rather than 'wait for element to appear'
While re-factoring automation for my project, i found that adhering to good practices resolves most of automation issues:
1. XPath: If IDs are available, use the same. Else, instead of relying on XPath constructed using HTML elements:
static final String DISPLAY_NAME = "//div[@id='body']/div[4]/div[1]/div/div[1]/h4"; static final String EMAIL_ADDRESS = "//html/body/div/div/div/div/span";
The following is more stable to minor changes in GUI:
static final String DISPLAY_NAME = “//div[@id='body']//div[@class='innerleft']/h4”; static final String EMAIL_ADDRESS = "//div[@id='global-info']/span";
Use @class, text(), contains etc, to create stabler Ids.
2. Instead of flaky, I prefer to use the “maintainability”. Sometimes, the debug information is less or confusing that does not let us to the actual issue quickly.
Ex:
20:04:58.539 INFO - Command request: isElementPresent[//a[text()='Remove image'], ] on session ece3c0c368e74594ab788751e13cea96
If the command fails identifying the element (timeout) , there is no clue regarding which page the above element belongs to.
Along with this, I suggest that we can have a high level log as below:
[UI Flow] Is Element Present >> EditProfilePage.RemoveImageLink
20:04:58.539 INFO - Command request: isElementPresent[//a[text()='Remove image'], ] on session ece3c0c368e74594ab788751e13cea96
3. There is always a temptation to add 'sleep(...)' in the code to get the automation running locally. I suggest any test code with sleep should not be allowed to be checked-in. Instead use 'waitForElementsToLoad(...);' to wait for actual element to be loaded before continuing with the automation.
Another practice which I found cleaner and also resolved issues is to keep the test scenario code away from the logic of waiting for a GUI element to load. This needs to be handled within the abstraction of the page itself.
Ex: (1) public void clickBloggerBuzz() { click(Ui.BUZZ_LINK); // Add the waitForElementsToLoad() here }
(2) Page constructors by default should wait for key element to be loaded before it returns.
public DashboardPage(Selenium selenium) throws MalformedURLException { super(selenium); waitForElementsToLoad(Ui.EDIT_PHOTO_LINK); }
Primarily, it describes how we got around having to assign hand-crafted IDs to all the Ext components (taking care of the fragile XPath problem V. Narayan Raman mentioned) and how we synchronise the tests with AJAX.
I work with Selenium since at least 3 years, and I have to admit it is quite buggy, forcing me to submit patches from time to time.
One of the problems of Selenium is that bugs are fixed very slowly, even when you submit a patch.
Another problem is that the code is not checked with an automated tool, like JsLint.
The most annoying problem is that Selenium randomly detects the loading of iframes. As Marius, I have a test suite that works fine on 2 computers, but fails on 2 others, and all computers are quite different. I'm still searching for the bug after having already spent a lot of time on it... I also found some problems related to Firefox, like the fact that retrieving window.document.body generates an exception sometimes !
Great post Simon. I think you hit the nail on the head. I will add to the list of poor test isolation with a few (fairly obvious) examples:
a) sharing data between tests b) running tests without proper cleanup c) running tests in parallel where current actions in one test might create unexpected outcomes in another
Writer is glossing over how buggy and unstable Selenium can be, especially for an automated build environment. As another commentor stated, simple calls to access elements will succeed one time but fail others.
Amen. I've run into any number of scenarios where the Selenium webdriver just ... stops. Everything looks fine, the browser is doing what it's supposed to be doing, but Selenium has somehow stopped interacting with the test runner. Running the whole thing again, the test that failed last time will work this time, but some other test will fail (or maybe, the whole batch will pass). Maybe it's my code, but I've never been able to isolate it. (And yeah, I'm following all the best practices he describes.)
Great Post. At the beginning of reading this post I had hope that I will find some suggestion/feedback/solution which will make my test suite stable. But still I’m finding myself in same point as I was before reading this post.
I agree with jcmeyrignac and Eric Jain comment.
Here is one of the latest bug from selenium guys: http://code.google.com/p/chromedriver/issues/detail?id=121, this clearly shows that there is no backward compatibility check before releasing selenium version.
If things don’t get stable, I guess soon there will be another open source testing tool to compete with selenium.
I agree with Marius Gedminas, Larry Port, & Xun. My primary cause of flakiness is waitForElement (or similar wait) succeeding and then the following action I take on the element fails to locate the element, even if I try to assure it is present & visible.
Same problem facing here too... Many test fails when we ran in jenkins but almost all test passes when we ran locally, No idea why it happens like that..
I am facing an issue while running selenium java code using webdriver. I tried in both linux and windows and both is throwing the same error. I tried to browse the solution in internet but most of the suggestion was to upgrade the selenium files. I am using the latest version of selenium and firefox. Below is the error am facing.Please help me resolve the issue.
org.openqa.selenium.firefox.NotConnectedException: Unable to connect to host 127.0.0.1 on port 7055 after 45000 ms. Firefox console output: Firefox.","creator":"Canonical Ltd.","homepageURL":null},{"locales":["sv-SE"],"name":"Ubuntu Modifications","description":"Ubuntu-paket för Firefox.","creator":"Canonical Ltd.","homepageURL":null},{"locales"
I have seen a Selenium test suite (for a very AJAXy application) that had random unexplainable nondeterministic failures. Actions like "wait for element with id=foobar to appear" and then the immediate next action "click on element with id=foobar" would suddenly say it couldn't find the element any more---but only sometimes and not always.
OdpovedaťOdstrániťPeople spent hours trying to find the root cause, without any luck.
With this type of error, particularly with an ajaxy application, you want to use 'wait for element to be clickable' rather than 'wait for element to appear'
OdstrániťYou missed XPath brittleness when UI changes with each dev build.
OdpovedaťOdstrániťHow do I ensure that with each DEV build, though my XPath changes, I can successfully run my tests?
OdstrániťDo you have any solution ?
"It's very seldom because you're uncovered a bug in the test framework."
OdpovedaťOdstrániťThat may be so, but browsing through their list of bugs turns up no small amount of unresolved "flakyness" issues.
Selenium is a great tool, but it's ironic that one of my testing tools should be the buggiest of all the dev tools I use...
While re-factoring automation for my project, i found that adhering to good practices resolves most of automation issues:
OdpovedaťOdstrániť1. XPath:
If IDs are available, use the same. Else, instead of relying on XPath constructed using HTML elements:
static final String DISPLAY_NAME = "//div[@id='body']/div[4]/div[1]/div/div[1]/h4";
static final String EMAIL_ADDRESS = "//html/body/div/div/div/div/span";
The following is more stable to minor changes in GUI:
static final String DISPLAY_NAME = “//div[@id='body']//div[@class='innerleft']/h4”;
static final String EMAIL_ADDRESS = "//div[@id='global-info']/span";
Use @class, text(), contains etc, to create stabler Ids.
2. Instead of flaky, I prefer to use the “maintainability”. Sometimes, the debug information is less or confusing that does not let us to the actual issue quickly.
Ex:
20:04:58.539 INFO - Command request: isElementPresent[//a[text()='Remove image'], ] on session ece3c0c368e74594ab788751e13cea96
If the command fails identifying the element (timeout) , there is no clue regarding which page the above element belongs to.
Along with this, I suggest that we can have a high level log as below:
[UI Flow] Is Element Present >> EditProfilePage.RemoveImageLink
20:04:58.539 INFO - Command request: isElementPresent[//a[text()='Remove image'], ] on session ece3c0c368e74594ab788751e13cea96
3. There is always a temptation to add 'sleep(...)' in the code to get the automation running locally. I suggest any test code with sleep should not be allowed to be checked-in. Instead use 'waitForElementsToLoad(...);' to wait for actual element to be loaded before continuing with the automation.
Another practice which I found cleaner and also resolved issues is to keep the test scenario code away from the logic of waiting for a GUI element to load. This needs to be handled within the abstraction of the page itself.
Ex:
(1)
public void clickBloggerBuzz() {
click(Ui.BUZZ_LINK);
// Add the waitForElementsToLoad() here
}
(2) Page constructors by default should wait for key element to be loaded before it returns.
public DashboardPage(Selenium selenium) throws MalformedURLException {
super(selenium);
waitForElementsToLoad(Ui.EDIT_PHOTO_LINK);
}
I love Selenium!
OdpovedaťOdstrániťWish I could get my head around this BDD though :(
I wrote an article on how we’ve been testing an ExtJS UI in Java at:
OdpovedaťOdstrániťhttp://www.xeolabs.com/portal/articles/selenium-and-extjs
Primarily, it describes how we got around having to assign hand-crafted IDs to all the Ext components (taking care of the fragile XPath problem V. Narayan Raman mentioned) and how we synchronise the tests with AJAX.
hope it helps somebody,
Lindsay
I work with Selenium since at least 3 years, and I have to admit it is quite buggy, forcing me to submit patches from time to time.
OdpovedaťOdstrániťOne of the problems of Selenium is that bugs are fixed very slowly, even when you submit a patch.
Another problem is that the code is not checked with an automated tool, like JsLint.
The most annoying problem is that Selenium randomly detects the loading of iframes.
As Marius, I have a test suite that works fine on 2 computers, but fails on 2 others, and all computers are quite different.
I'm still searching for the bug after having already spent a lot of time on it...
I also found some problems related to Firefox, like the fact that retrieving window.document.body generates an exception sometimes !
Great post Simon. I think you hit the nail on the head. I will add to the list of poor test isolation with a few (fairly obvious) examples:
OdpovedaťOdstrániťa) sharing data between tests
b) running tests without proper cleanup
c) running tests in parallel where current actions in one test might create unexpected outcomes in another
Writer is glossing over how buggy and unstable Selenium can be, especially for an automated build environment. As another commentor stated, simple calls to access elements will succeed one time but fail others.
OdpovedaťOdstrániťAmen. I've run into any number of scenarios where the Selenium webdriver just ... stops. Everything looks fine, the browser is doing what it's supposed to be doing, but Selenium has somehow stopped interacting with the test runner. Running the whole thing again, the test that failed last time will work this time, but some other test will fail (or maybe, the whole batch will pass). Maybe it's my code, but I've never been able to isolate it. (And yeah, I'm following all the best practices he describes.)
OdstrániťAs another commentor stated, simple calls to access elements will succeed one time but fail others.
OdpovedaťOdstrániťCannot agree more! Selenium creates more problems, produces more false alerts that I have to deal with than the actual coding.
Great Post.
OdpovedaťOdstrániťAt the beginning of reading this post I had hope that I will find some suggestion/feedback/solution which will make my test suite stable.
But still I’m finding myself in same point as I was before reading this post.
I agree with jcmeyrignac and Eric Jain comment.
Here is one of the latest bug from selenium guys: http://code.google.com/p/chromedriver/issues/detail?id=121, this clearly shows that there is no backward compatibility check before releasing selenium version.
If things don’t get stable, I guess soon there will be another open source testing tool to compete with selenium.
I agree with Marius Gedminas, Larry Port, & Xun. My primary cause of flakiness is waitForElement (or similar wait) succeeding and then the following action I take on the element fails to locate the element, even if I try to assure it is present & visible.
OdpovedaťOdstrániťDid you ever find a solution to this?
OdstrániťSame here, I can run a test and it fails to find element, yet it really finds it and test succeeds at the end, yet always has those failures.
OdpovedaťOdstrániťIs there anyway to tell selenium to ignore those failures, when running as a htmlsuite?
Did you by chance find a solution to this, we are experiencing something similar and just cannot resolve...
OdstrániťSame problem facing here too... Many test fails when we ran in jenkins but almost all test passes when we ran locally, No idea why it happens like that..
OdpovedaťOdstrániťHi All,
OdpovedaťOdstrániťI am facing an issue while running selenium java code using webdriver. I tried in both linux and windows and both is throwing the same error. I tried to browse the solution in internet but most of the suggestion was to upgrade the selenium files. I am using the latest version of selenium and firefox. Below is the error am facing.Please help me resolve the issue.
org.openqa.selenium.firefox.NotConnectedException: Unable to connect to host 127.0.0.1 on port 7055 after 45000 ms. Firefox console output:
Firefox.","creator":"Canonical Ltd.","homepageURL":null},{"locales":["sv-SE"],"name":"Ubuntu Modifications","description":"Ubuntu-paket för Firefox.","creator":"Canonical Ltd.","homepageURL":null},{"locales"