to do not lost pieces abruptly, since the held amount is not enough incentive for many SNO to perform a stressful GE, like it was implemented.
If the exiting node would be online and have all needed pieces to perform a repair (which works stable and without such stress for both sides), this would help the network to keep the specified repair threshold without a risk to lost pieces because of abruptly shutdown of nodes.
the current implementation of GE is unable to help with that case anyway - your node will exiting at least a month anyway due to many limiting factors, include slow and unstable upstream bandwidth on the nodeās side and the high load on the satellites to prepare list of nodes for each exiting node in many loops (because your node is unable to transfer pieces from the first attempt for any reason, so it requests 4 more nodes to attempt to transfer for each piece).
yes, itās a cost to receive your held amount, but now without a stress and risk of DQ during the process.
Thatās why if You are revamping, do it for a better,
And my upstream is not slow,
1 node i had, i exited in 2 days out of 1 TB data it has, so if You are revamping make it complete as soon its done, You can pay in 1-2 months no problem, just donāt expect me to keep the node online for another 3-4 weeks. All im saying is, if You are revamping, take the opportunity to do it right. Also jammerdan said very important for SNOs thing, partial GE, to choose how much data cut, without exiting.
The slow and/or unreliable upstream mean that node is unable to upload pieces without a timeout to other nodes. It could be very OK to upload a one piece/5 minutes, but not thousands in parallel. Highly depends on your router to be able to process this amount of parallel transfers both ways simultaneously (the ingress is not stopped in the current implementation).
So, your upstream bandwidth could be fast and reliable to the nearest datacenter within the same ISP, but incredibly slow and packets-droppy to the next country.
The partial GE is possible to achieve with a dangerous offline on the edge balancing to be DQ - the more your node offline, the more data it will lost but greatly increases the chance to be DQ. So the safest method of partial GE is set the storage allocation below the usage. Or exit from one or several satellites (decreasing storage allocation in the same time, otherwise this free space will be used by the customers of remained satellites).
The current implementation doesnāt have even a possibility to do a partial GE, so to achieve that it should be re-designed.
Is the new GE system active? I just GE on 9 nodes from Saltlake, and the younger ones with aprox 250-350 GB finished in a day; the older ones with more than a TB still working.
I started a GE August 23rd, itās been over a month and only at 33% on US1. Will my node ever show that it is gracefully exited? Iām getting pressured from IT to shut down this system and I had no idea that a graceful exit was a 3 month exit process.
Just keep an eye on it after that, if you would see a lot of transfer errors, you need to reduce it, otherwise your node could be disqualified if the number of failed transfer would be greater than 10%. Each failed piece transfer will be retried 4 more times before considered as failed.
The instruction have not yet been updated to reflect the merged changes. This is still the most current published instruction but we have a request in to the team for updated instruction and will post in the Forum when they are available.