Welcome to Communities.
"How many hosts can be in maintenance mode in the same time without have risk?"
This depends entirely on what you mean by "risk" - generally I would consider data being at risk if it has no redundancy for a prolonged period.
What is the intended purpose of placing the nodes in Maintenance Mode?
For example, are you performing updates/patches that can be done one node at a time or is it some site maintenance that requires an entire site shut down and/or for a relatively prolonged period (e.g. 10+hrs/days)?
"I know that all hosts of one site could be in maintenance mode but if one host from the other site will fail,could have we problems?"
So, with a Storage Policy of PFTT=1, SFTT=1 LocalFTM=RAID5, this is essentially a RAID1 (across sites) of RAID5s - this means an entire site can be shut down/in MM and the data will still be FTT=1 on the remaining site, this means the data will still be available if a subsequent single node/disk/Disk-Group failure occurs (though then it will be FTT=0 until it has repaired it back to FTT=1).
"Whats is it the best practices for maintenance mode in Stretched Cluster? only one host in the same time?"
As I alluded to above - it depends what you are doing - if a whole site needs to be shut down for maintenance, then this is the very reason you have PFTT=1 and SFTT=1, but also if you wanted to update a whole site at a time then this is fine provided your data is adequately protected and you understand the implications of doing it this way as opposed to one at a time.
The main reason is performing update of firmware, drivers and the upgrading the esxi version. So, I would like to do this secure manner and the shortest time possible. I used to do one host in the same time but maybe there is other faster way.
Thank you very much for you answer.
Provided all the data is indeed using a Storage Policy with dual-site local-protection then patching one site at a time (and don't forget the Witness!) should be fine - I see customers doing this regularly without issues but as with any maintenance (on vSAN or otherwise), ensure you have a good and current set of backups before you begin.
I would advise confirming that all Objects have a dual-site policy by looking at the data generated from 'esxcli vsan debug object list' (with flag '--all' if on 6.7 U3 or later) - if you PM me this I can take a look.