Whilst I wasn't the only one voting to close - it took four of my colleagues to agree with me - I was the first person to so tag it, so I think it behooves me to offer some kind of explanation.

Your original question asked about how to prioritise processes so that the OOM killer would kill them in the order you desired. That has been answered, in the duplicate question pointed to. To the best of my knowledge, the technique described can be used to do all three of the substantive tasks you raised in your original question: make a process unlikely to be killed, make one very likely to be killed, and sort them in order of killing priority (which subsumes questions one and two, anyway).

Now, you might already know all that. You might have done a bunch of work with oom_score_adj and found that it doesn't offer you quite the control you have a clearly-communicable business need for, or established that in practice it doesn't work quite as described, and you need help tuning it. But you don't tell us about any research or work you've already done. You don't indicate that you've done any searching on SF, or anywhere else, already. You give us no reason to believe that oom_score_adj is anything other than the right tool for the job.

So either your question means what it said, in which case it's been answered already; or there's more to your question than meets the eye, but you haven't put in the extra work to make your question communicate that. If it's the latter, then expecting us to drag it out of you one comment at a time is just plain rude.

Edit: you asked "how can I do X", and have conceded that "use Y" was a "valuable answer". If that answer would have been enough for you to accept it, the question is a duplicate, because the answer to the one is the answer to the other. If it wouldn't, could you clarify what else you wanted from the answer?

Whilst I wasn't the only one voting to close - it took four of my colleagues to agree with me - I was the first person to so tag it, so I think it behooves me to offer some kind of explanation.

Your original question asked about how to prioritise processes so that the OOM killer would kill them in the order you desired. That has been answered, in the duplicate question pointed to. To the best of my knowledge, the technique described can be used to do all three of the substantive tasks you raised in your original question: make a process unlikely to be killed, make one very likely to be killed, and sort them in order of killing priority (which subsumes questions one and two, anyway).

Now, you might already know all that. You might have done a bunch of work with oom_score_adj and found that it doesn't offer you quite the control you have a clearly-communicable business need for, or established that in practice it doesn't work quite as described, and you need help tuning it. But you don't tell us about any research or work you've already done. You don't indicate that you've done any searching on SF, or anywhere else, already. You give us no reason to believe that oom_score_adj is anything other than the right tool for the job.

So either your question means what it said, in which case it's been answered already; or there's more to your question than meets the eye, but you haven't put in the extra work to make your question communicate that. If it's the latter, then expecting us to drag it out of you one comment at a time is just plain rude.

Edit: you asked "how can I do X", and have conceded that "use Y" was "a valuable answer". If that answer would have been enough for you to accept it, the question is a duplicate, because the answer to the one is the answer to the other. If it wouldn't, could you clarify what else you wanted from the answer?

Whilst I wasn't the only one voting to close - it took four of my colleagues to agree with me - I was the first person to so tag it, so I think it behooves me to offer some kind of explanation.

Your original question asked about how to prioritise processes so that the OOM killer would kill them in the order you desired. That has been answered, in the duplicate question pointed to. To the best of my knowledge, the technique described can be used to do all three of the substantive tasks you raised in your original question: make a process unlikely to be killed, make one very likely to be killed, and sort them in order of killing priority (which subsumes questions one and two, anyway).

Now, you might already know all that. You might have done a bunch of work with oom_score_adj and found that it doesn't offer you quite the control you have a clearly-communicable business need for, or established that in practice it doesn't work quite as described, and you need help tuning it. But you don't tell us about any research or work you've already done. You don't indicate that you've done any searching on SF, or anywhere else, already. You give us no reason to believe that oom_score_adj is anything other than the right tool for the job.

So either your question means what it said, in which case it's been answered already; or there's more to your question than meets the eye, but you haven't put in the extra work to make your question communicate that. If it's the latter, then expecting us to drag it out of you one comment at a time is just plain rude.

Whilst I wasn't the only one voting to close - it took four of my colleagues to agree with me - I was the first person to so tag it, so I think it behooves me to offer some kind of explanation.

Your original question asked about how to prioritise processes so that the OOM killer would kill them in the order you desired. That has been answered, in the duplicate question pointed to. To the best of my knowledge, the technique described can be used to do all three of the substantive tasks you raised in your original question: make a process unlikely to be killed, make one very likely to be killed, and sort them in order of killing priority (which subsumes questions one and two, anyway).

Now, you might already know all that. You might have done a bunch of work with oom_score_adj and found that it doesn't offer you quite the control you have a clearly-communicable business need for, or established that in practice it doesn't work quite as described, and you need help tuning it. But you don't tell us about any research or work you've already done. You don't indicate that you've done any searching on SF, or anywhere else, already. You give us no reason to believe that oom_score_adj is anything other than the right tool for the job.

So either your question means what it said, in which case it's been answered already; or there's more to your question than meets the eye, but you haven't put in the extra work to make your question communicate that. If it's the latter, then expecting us to drag it out of you one comment at a time is just plain rude.

Edit: you asked "how can I do X", and have conceded that "use Y" was a "valuable answer". If that answer would have been enough for you to accept it, the question is a duplicate, because the answer to the one is the answer to the other. If it wouldn't, could you clarify what else you wanted from the answer?