On December 19, 2005, Helix Server 6.0 shipped and the beta phase came to a close.
Links to the current information are found above. Links documenting the process are found to the left. This page remains as a testament to the work that went into it.
|Final Candidate Test Information
Update: Starting on Decenber 1st, this test uses build 6.0fc2(057). This build breaks compatibility with all builds prior to 6.0fc1 (054), so if you downloaded an earlier build be sure to delete it (and all earlier Kapi Clients) and download 6.0fc2(057).
The Final Candidate Test began November 15th and continues until we ship Helix 6.
The Friday chat is now moved back to the regular server, qsatoolworks.com where it will remain online most of the time. (Note please, that it will be down from time to time as we work in RADE to add features and integrate our new and improved account management site.
Our first 4 weeks of "final candidate" testing turned up a couple of problems which have been addressed. Unless something unexpected turns up, this build will be "one shy" of the actual release verison of Helix. (The the only changes in the release build are to set the version number to "6.0".)
Be sure to keep an eye out for our release date announcement, coming soon.
|What's Being Tested
Pele Server/Kapi Client: Our current tests are as follows: we are running a macOS Native (Pele) Server here, and we invite you to visit using an OS 9 Native (Kapi) Client.
This will tell us two things:
- Pele Server is solid and can handle large numbers of Clients visiting all at once.
- We can successfully cross the chasm between an OS 9 Client and a macOS Server.
So for now, we are asking you to download and test the OS 9/Classic compatible version of Helix Client 6.0. Although it may be a bit disappointing to realize that it will look exactly the same on your end as it does with Helix Client 5.3, the realization that your Client is really talking to a Helix Server running native macOS code should be good news.
If you encounter problems, please report them through the Defect Report Entry form in the Bug Reports hierarchial menu under the Guest menu.
Some of you are noticing the obvious Catch-22: what if you can't even visit the new Server to file the report? If that's the case, please visit our contact web form and send a message to Matt Strange. Specify "6.0 Preview Problem" in the Subject field.
Only by using one of those methods can we guarantee your bug report will not get lost. If you call or email one of us directly, we might lose the note (or garble the meaning) and it won't get handed to the correct engineer intact. By reporting via the requested methods you can be assured that the engineer who works on the bug will see it in your own words and will be able to contact you directly, should that be necessary.
But of course, if neither of those works for you, we want your report, so use whatever method you know to get in touch with us.
To all who are able to help us out during this critical time, we say thank you! Our nerves are frazzeled and we sometimes may appear unresponsive (or worse, cranky from lack of sleep) but please understand that we really do appreciate your assistance.
|Code Names 101
The next release on our schedule is Helix 6.0.
Lest anybody be confused, please note: The "Version 6.0" identifier does not mean the product is "macOS native." Some Helix 6.0 products will run in only macOS, some will run only in OS 9 and some will run in both. Helix 6.0 products running in OS 9 will be labelled Helix 6.0 Classic as a reminder.
Helix 6.0 goes by two codenames: Pele is the code name for our macOS native branch, and Kapiolani (Kapi for short) is the code name for the OS 9 native branch. Please use these code names when communicating with us about these products.
Helix 5.3 (Chaski) was built speficially to get the totally rewritten, macOS native TCP/IP code into your hands sooner. Since that was the major piece of code that had to be reworked for the Pele project, the most critical code rewrite has been in your hands and very well tested over the course of the last year. Now we are ready to move on.
It is our plan to release each Helix 6.0 component as soon as it is ready. Based on our current expectations, our first Helix 6.0 release will break down as follows:
- macOS Native: Helix Server
- OS 9 (Classic) Only: Helix RADE, Client, and Engine
- OS 9/X Native: Update Collection, Helix Utility
As soon as Helix Server 6.0 ships, our attention will turn to Helix Client, then Engine, and finally RADE. For more information on this phased release plan, please see this announcement from April of 2004.
|Test 1 Results
The first test picked up where we left off with the Chaski testing: the HelixChat. The intent was to have as many people as possible connect at one time.
Guests did their level best to stress the new server but it held up very well. After about 4 hours there were 129 clients visiting. The performance decline seen with that many Clients was noticable, but not extreme, and to be expected.
However, shortly after the 129th Client logged in, the Server crashed. Diagnositcs showed that the test collection suffered no damage.
The crash log did not yield detailed data, but our current speculation is that there is a single byte counter that is treated as a signed integer, and when it exceeds 127, it "goes negative" which can cause all sorts of bizarre types of crashes.
During our next test, we hope to be able to follow the Client count closely to determine whether this is the case or not. If it proves to be the source of the crash, we will defer this bug until a later date, as our current licenses cap at 100 Clients. (We look forward to the day somebody purchases a Server licensed for 128 Clients, so we can revisit this one!)
Thank you once again to all who particpated in this round of testing.
|Test 2 Results
Test two was held June 24th.
We put some real stress on the Server, having multiple users posting groups of records simultaneously while others had lists open that were updated with the posted data. The sheer volume of work brought the Server down three times in the span of a couple of hours.
Based on examination of the crash logs, we have determined that the crashes were due to the extreme stress we put on the Server, and that the same crash would occur in earlier versions as well. The bug appears to be very deep in the code and will take extensive work to fix. As such, we are going to defer work on this specific bug until that section of the code is up for rewrite. (Of course, if wider beta testing shows this not to be the case, we will certainly address it prior to release.)
Thank you once again to all who participated.
|Test 3 Results
Test three was held July 1st.
This week's test was intended to be a simple "sanity check" test before we release the beta to testers, but some bizarre events caused us to delay releasing a beta for another week. We hoped to get a large number of visitors to test the alleged 128 user cap, but we crashed before we could get that high.
During the test we had numerous reports of Clients getting "the server has vanished" messages and we experienced two Server crashes. After studying the crash logs, we have come to the tentative conclusion that July 1 was a very flaky day for the internet in general and that may have affected our test. We decided to try again with essentially the same code next week and see if the problems repeated themselves.
|Test 4 Results
Test four was held July 8th.
This test was a repeat of last week's "sanity check" test. Although we did not see the same crash on the Server, we did see some instances of Clients being terminated for no apparent reason. Perhaps Helix is a little to finicky about internet connection speed and any abnormal delay is interpreted as a "lost" Client. We'll have to investigate that further.
We did experience one crash, which our engineers said is identical to the ones we experienced in test two. Further examination shows that these crashes are apparently being caused by problems with the code that handles document streams. Although we already knew that code was not completely stable, and are planning on overhauling it once we get the macOS products shipped (which is better than the 5.x counterpart even with this bug) we are going to work on a "bandaid" to make it less troublesome.
The test server will most likely remain online until Thursday July 14, when we will be putting up next week's build. (As always, the "non-test" chat at qsatoolworks.com using the Helix 5.3.2 Client is available throughout the week.)
|Test 5 Results
Test five was held on July 15th.
This test was a repeat of the basic tests along with the testing of the new 'server status' icons in the Server Info window.
The test went well. Bugs were found, feedback gathered, and Helix refined as a result.
|Test 6 Results
Test six is on July 22nd and July 29th.
This test features the new and improved "6.0b2" code. Lots of bug fixes have been put into place.
We found one interesting bug that resulted in a mini-test on Wednesday, Aug 3rd. The result of that test were encouraging, but we still have work to do.
|Test 7 Results
Test seven ran August 5th through August 18th.
This test featured the new and improved "6.0b3" code. Lots of bug fixes had been put into place.
The first goal of this release was to aggressively chase the crashes that hit the Server during our weekly chat. In order to pursue this, we turned on a number of debugging aids.
During this two week test phase, we experience no crashes at all. Whether this is attributable to the improved code, the debugging aids we turned on (by slowing the performance, we may have cloaked the problem), or some other undetected change. Our next test will start with the non-debugging Server to see if the problem returns.
|Test 8 Result
Test eight ran from August 19th through Sept 7th.
This test used the 6.0b4 code. Many more bug fixes have been put into place, and some of the rough edges are starting so show polish.
The Server crashes are completely gone now, so that is a postive.
We were getting scattered reports that the chat window was not displaying every update. Some Client would see new messages in the "Chat Review" window but not in the live chat window. With subsequent testing we were able to isolate this bug. We are working on a fix, but as of Sept 8 none had been found.
|Test 9 Results
Test nine started September 9th and ended September 12.
This test used build 6.0b5, which has been in the hands of beta testers since Sept 2nd.
This test was cut short because the fix for the lost records bug was found. We are now moving immediately to build 6.0b6. See Test 10 Info.
|Test 10 Results
Test ten ran September 13th. It used an early release of build 6.0b7, but was quickly reverted back to 6.0b6 when a problem cropped up that resulted in multiple Server freeze ups.
That problem appears to be fixed, and the "real" 6.0b7 test will begin on Sept 23rd.
In the meantime, we've knocked off a whole bunch of other little bugs, and we're starting to think we are getting close to the end of the tunnel. We'll know when Test 11 runs.
|Test 11 Results
Test eleven ran September 23-28.
This test uses build 6.0b7 (043), which replaces the short-lived 6.0b7 (041).
Recent tests isolated a situation where a Client whose network pipe was saturated would quietly ignore messages, failing to display all records in a list. This test showed that bug is fixed.
|Test 12 Results
Test twelve ran September 30 - Oct 7.
This test used build 6.0b8 (045) which attemped to prove that our changes to the Client disconnect code were solid. Unfortunately, some testers logged in and walked away, leaving their Clients logged in beyond the end of the test time. This made it impossible to test what we hoped to test.
We would like to run this test again, but other important fixes have been made, so we are going to bump up to the next beta and try to test two sets of code changes at once.
|Test 13 Results
Test thirteen was scheduled for October 14th, but a hardware failure (complete with kernel panic) derailed our test. This test was rescheduled for October 21st and ran successfully from then until it was taken down around the end of the month.
This test used build 6.0b9 which includes a significant bug fix related to AND/OR query evaluation. Unfortunately, that fix resulted in another problem cropping up, so we went back to the drawing board. 6.0b10 addresses this new bug.
|Test 14 Results
Test fourteen ran from November 4th through November 15.
This test used build 6.0b10 which is very close to the final version.
Test 14 included the "fix for the fix" noted in Test 13 (see above) and also marks the first beta release where we have addressed every bug reported by our beta testers to date. This test proved that we are rapidly approaching the finish line.
This test marks the end of the test server database. Thank you to all who helped us.