*: update build process and ssh in bench.py

[martingrambow]

There were some small errors in bench.py (fixes #1202):

• ssh_connect_with_retries() did not ask for the ssh key
• I also added some logging in ssh_cmd()
• bootstrap(addr) followed the old build process
• run() called the async ssh command, however, sometimes the pkill killed the deamon immediately; so I switched to the usual ssh_cmd and to screen

[ploxiln]

Clearly we have not maintained this script in the last couple years. Thanks.

[ploxiln]

while you're here, the defaults for the instance_type and ami options should probably be updated

[martingrambow]

Thanks for the feedback, I think I fixed all the issues.

[ploxiln]

This ami does not seem to exist in the default region, us-east-1 (and I also checked a couple of other common regions and could not find it). Can you find an ami from the default region?

$ aws --region us-east-1 ec2 describe-images --image-ids ami-48fd2120
{
    "Images": [
        {
            "Architecture": "x86_64",
            "CreationDate": "2014-08-16T21:40:48.000Z",
            "ImageId": "ami-48fd2120",
            "ImageLocation": "ubuntu-us-east-1/images/hvm-instance/ubuntu-trusty-14.04-amd64-server-20140816.manifest.xml",
            "ImageType": "machine",
            "Public": true,
            "OwnerId": "099720109477",
            "State": "available",
            "BlockDeviceMappings": [],
            "Hypervisor": "xen",
            "Name": "ubuntu/images/hvm-instance/ubuntu-trusty-14.04-amd64-server-20140816",
            "RootDeviceType": "instance-store",
            "SriovNetSupport": "simple",
            "VirtualizationType": "hvm"
        }
    ]
}
$ aws --region us-east-1 ec2 describe-images --image-ids ami-02df9ea15c1778c9c

An error occurred (InvalidAMIID.NotFound) when calling the DescribeImages operation: The image id '[ami-02df9ea15c1778c9c]' does not exist

(I guess we should update to ubuntu-16.04 or ubuntu-18.04, but you probably did that, I just can't tell because I can't find the ami ...)

[ploxiln]

I think c4.xlarge or c5.xlarge would be more in the spirit of the previous default. A bit expensive, but they'll run for less than an hour, and t2 or t3 instances are probably not good for benchmarking because they are more oversubscribed and throttled.

[ploxiln]

This part did not have to change. The restructure is OK with me, except that using parentheses like this is odd and unusual for python (they're not needed).

[mreiferson]

see #1264