Musk: V10 wide release "hopefully end of august" after early access

[Silicon Desert]

That's a common practice in the industry.

The way to mitigate big disaster is to release slowly and stop the roll out if you find issues, fix the issue - rollout new version. That is exactly what they are doing.

ps : BTW, its not really new either. We were doing frequent releases and monitored slow roll out 20 years back. For eg., Amazon probably makes some kind of release every day. Hasn't resulted in disasters.

yea, you beat me to it. I've been doing somewhat similar methods for as long as 30 years. I used to do a big beta test before rolling out any change.

 

[motocoder]

That's a common practice in the industry.

The way to mitigate big disaster is to release slowly and stop the roll out if you find issues, fix the issue - rollout new version. That is exactly what they are doing.

ps : BTW, its not really new either. We were doing frequent releases and monitored slow roll out 20 years back. For eg., Amazon probably makes some kind of release every day. Hasn't resulted in disasters.

Yeah, i know all about incremental rollout, canary deployments, etc. I work in Microsoft Azure, and before that I worked in Bing were we did deloyments over hundreds of thousands of servers. But all of that you're deploying something that's had some internal testing first, and less than 1 day is not a realistic amount of time for that.

 

[Silicon Desert]

But all of that you're deploying something that's had some internal testing first, and less than 1 day is not a realistic amount of time for that.

Oh wait a second there. It seems your assumption is the changes were done in less than a day. Not usually the case with these rollouts. 2019.32.11 can be something as simple as including a minor improvement to the code base of .10 being worked on a week earlier and just released, and may not even have anything to do with driving or other important tasks. We've seen that on many releases.

 

[motocoder]

Oh wait a second there. It seems your assumption is the changes were done in less than a day. Not usually the case with these rollouts. 2019.32.11 can be something as simple as including a minor improvement to the code base of .10 being worked on a week earlier and just released, and may not even have anything to do with driving or other important tasks. We've seen that on many releases.

It just seemed like sort of a panic fix rollout to me. If this release was ready, why push out 2019.32.10 the day before?

 

[kaffine]

Because at some point you have to do a release. If you keep the well we have another bug fix that is going to be done testing tomorrow you will never actually release as you will always have another bug fix coming out soon.

 

[motocoder]

Because at some point you have to do a release. If you keep the well we have another bug fix that is going to be done testing tomorrow you will never actually release as you will always have another bug fix coming out soon.

well, we may never know what the motivation was. In any event, I am loving V10.

 

[EVNow]

Yeah, i know all about incremental rollout, canary deployments, etc. I work in Microsoft Azure, and before that I worked in Bing were we did deloyments over hundreds of thousands of servers. But all of that you're deploying something that's had some internal testing first, and less than 1 day is not a realistic amount of time for that.

Yes - before all that we used to do the same with MSN/Passport/XBox etc. (I've moved on and don't work at MS anymore).

But now its not uncommon to have daily patches - and ofcourse weekly patches. We do that in my current company. Probably not in Azure or O365, yet.

I'm sure - like most places - Tesla has an automated test suit they will run through before releasing.

ps : I should have figured out you are in software by looking at the alias

 

[Silicon Desert]

It just seemed like sort of a panic fix rollout to me. If this release was ready, why push out 2019.32.10 the day before?

yea, it sometimes seems like a panic patch change to folks outside of the process. It is probably a natural assumption that a new release coming out so quickly is due to something important and suddenly discovered, yet rarely the case.

I won't speculate to make excuses for Tesla, but understand many ways this can happen. In the past when I have done engineering on large projects, I had several different teams working on different software features. A couple of my clients have used a process similar to Tesla (Julian date release numbers), but not something I prefer. One team comes up with changes they wanted to include which were totally unrelated to other features recently released, so they merge their changes in with the most recent release and then pushed it out.

I'm not going to second guess Tesla's process. From the relatively few developers I know over here, they are VERY SMART, and many of them a lot smarter than me although I've been doing software/electronics engineering for 40 years. Some of those guys think at 80 miles an hour while my brain is like the current smart summon in a parking lot

 

[EVNow]

It just seemed like sort of a panic fix rollout to me. If this release was ready, why push out 2019.32.10 the day before?

One of the old hands used to tell us (@MS) that if something is difficult, do it often. It gets easy

Tesla certainly does a lot of releases, and often.

In case you have not seen … TeslaFi.com Firmware Tracker

 

[conman]

2 hour drive on autopilot on the freeway today and it was exceptional! Smooth accelerating and braking, no phantom braking, no microbraking for adjacent vehicles, no surprise abrupt slowdowns when cars merge in front, very smooth lane changes. Visualisation even shows cars going in the opposite direction and have new icons - like pick-up trucks. We might have not gotten any new AP features, but existing AP is the best it's ever been by far.

 

[EVNow]

Here are some of my observations :
- No change in city roads with no center lane.
- In one particularly difficult curvy intersection, the car would go into left turn lane of oncoming traffic. Same now.
- Hesitates less and sticks to the lane better, at places where the lane splits
- The visualization of traffic in the other direction is good - but doesn't show all the cars. Misses some.

 

[Saghost]

Here are some of my observations :
- No change in city roads with no center lane.
- In one particularly difficult curvy intersection, the car would go into left turn lane of oncoming traffic. Same now.
- Hesitates less and sticks to the lane better, at places where the lane splits
- The visualization of traffic in the other direction is good - but doesn't show all the cars. Misses some.

My theory, which I don’t have enough data to flesh out yet, is that the oncoming traffic is only displayed in the cluster for certain road configurations.

Or did you have back to back cars on the same stretch of road with no changes mix displaying and not?

My limited experience didn’t have that - on a given stretch, I think all showed or none did.

 

[bhakan]

"Come to me" smart summon does not work for me. Only once I could have the car come to me in an empty parking lot and the car came forward. However, the summon failed when I tried last night and this morning this "come to me' and asked the car to come to me in reverse where I was standing because that is how I park in my driveway, back in first. So, parking in reverse or come to me in reverse and turning and come backward did not work.

 

[EVNow]

Or did you have back to back cars on the same stretch of road with no changes mix displaying and not?

My limited experience didn’t have that - on a given stretch, I think all showed or none did.

This was a 2 lane road @ 60 mph. The traffic was steady but not back to back. There was no pattern to which cars were displayed and which were not.

 

[motocoder]

One of the old hands used to tell us (@MS) that if something is difficult, do it often. It gets easy

Tesla certainly does a lot of releases, and often.

In case you have not seen … TeslaFi.com Firmware Tracker

yes, i’m aware of the firmware tracker. I thought of another reason they might roll out two releases back-to-back like that. Sometimes you have some code that only needs to be there temporarily during some transition, and then once the transition is complete, you deploy another update without that code (to reduce the memory footprint, simplify the code base, etc.). For example, suppose in version V10 they changed the format of some file they used to store settings and there was some code in the first V10 release, 2019.32.10, to migrate from the old format to the new. Then once the car reports in indicating it's running 2019.32.10 you deploy another release without that code, 2019.32.11, to that car because the conversion code is for sure not needed anymore.

 

[bhakan]

The latest V10 update identified very nicely cars turning left in to their respective lanes when it is left turn for both directions at a traffic light when I am sitting at the red light. As you can see there is no AP as my car is the lead car and I am traveling straight waiting for my green light. First time I have seen this. See the attachment:

 

[Saghost]

yes, i’m aware of the firmware tracker. I thought of another reason they might roll out two releases back-to-back like that. Sometimes you have some code that only needs to be there temporarily during some transition, and then once the transition is complete, you deploy another update without that code (to reduce the memory footprint, simplify the code base, etc.). For example, suppose in version V10 they changed the format of some file they used to store settings and there was some code in the first V10 release, 2019.32.10, to migrate from the old format to the new. Then once the car reports in indicating it's running 2019.32.10 you deploy another release without that code, 2019.32.11, to that car because the conversion code is for sure not needed anymore.

The most interesting part to me is that 32.10.x is forked - 10.1 only to 3s, 10.2 only to S/X - but the replacement 32.11 appears to be the same universal version for both as has been typical.

 

[EVNow]

For example, suppose in version V10 they changed the format of some file they used to store settings and there was some code in the first V10 release, 2019.32.10, to migrate from the old format to the new. Then once the car reports in indicating it's running 2019.32.10 you deploy another release without that code, 2019.32.11, to that car because the conversion code is for sure not needed anymore.

No - there are people now getting updates directly from 32.2 to 32.11.

The most interesting part to me is that 32.10.x is forked - 10.1 only to 3s, 10.2 only to S/X - but the replacement 32.11 appears to be the same universal version for both as has been typical.

Yes - 32.11 seems to be a combined version for all models.

 

[S100D]

I am on 2019.32.11 and just noticed two bugs:

1. My Internet connection keeps dropping every 20 seconds.
   
2. I used AP to navigate to the urban SC and with an Internet connection, I got a message that Navigate on Autopilot was not available. ‘‘This was on a road where it had been available before.
3. I also lost the ability to give voice commands. I tried to enter a bug report and navigate to a location.

All of the above was working earlier this morning.

Edit: I rebooted and it the Internet is not dropping now, but it seems to have fewer bars than before I installed 32.11.

I was not on a road after charging where I could check Navigate on Autopilot.

After rebooting, I was able to give voice commands.

 

[motocoder]

The most interesting part to me is that 32.10.x is forked - 10.1 only to 3s, 10.2 only to S/X - but the replacement 32.11 appears to be the same universal version for both as has been typical.

That is interesting.