1

Topic: and storage

I will outline also why  rises  when storage comes to an end? At Windows of the such did not watch better to hang up than  the last process?

2

Re: and storage

Hello, Vanja Pervachev, you wrote: > I will outline also > why  rises  when storage comes to an end? > at Windows of the such did not watch On my observations - all strictly on the contrary, on one almost  Linux from shortage of storage - 10-20 of full  at Windows. > it is better to hang up than  the last process? What of them last?

3

Re: and storage

Hello, netch80, you wrote: N> On my observations - all is strict on the contrary, on one almost  Linux from shortage of storage - 10-20 of full  at Windows. Can simply storages was much and had not time to come to an end? >> it is better to hang up than  the last process? N> what of them last? Last which launched also storage came to an end

4

Re: and storage

Hello, netch80, you wrote: N> On my observations - all is strict on the contrary, on one almost  Linux from shortage of storage - 10-20 of full  at Windows. On my observations just at  deadly problems with shortage of storage - was  on which often happened OOM Kill (on a host there was no enough storage that it to increase), and the result of it always murder of the same sshd - to be connected on a network it is impossible more, and the input through the console too ceases to work - only hard . The Windows all do not shoot to itself at a head at shortage of storage.

5

Re: and storage

Hello, Vanja Pervachev, you wrote: > I will outline also > why  rises  when storage comes to an end? > at Windows of the such did not watch How much I  Windows copy storage in a swap when it some time does not change, without releasing thus storage page. And when storage comes to an end, the pages which copies are saved in the swap file simply are released. At , like, pages are preempted swapping already at shortage of storage (though in modern  I will not be guaranteed, but like a principle remained the same) that not bad brakes system. > it is better to hang up than  the last process? OOM kill in  does both that, and another

6

Re: and storage

> I will outline also > why  rises  when storage comes to an end? > at Windows of the such did not watch Because at Windows one of standard tests app/driver verifier' - operation in the conditions of shortage of storage. And in  shortage of storage always . However, on another to be and cannot, with its lazy strategy  physical pages,  here simply formalization de facto behavior.

7

Re: and storage

Hello, Somescout, you wrote: S> On my observations just at  deadly problems with shortage of storage - was  on which often happened OOM Kill It from that " !", and then  forget to add in process of loading increase.

8

Re: and storage

Hello, Somescout, you wrote: N>> On my observations - all is strict on the contrary, on one almost  Linux from shortage of storage - 10-20 of full  at Windows. S> On my observations just at  deadly problems with shortage of storage - was  on which often happened OOM Kill (on a host there was no enough storage that it to increase), and the result of it always murder of the same sshd - to be connected on a network it is impossible more, and the input through the console too ceases to work - only hard . The Windows all do not shoot to itself at a head at shortage of storage. A swap to include, oom_score_adj to the necessary processes to expose. Normally system init it supports (at least upstart and systemd). Well also process is killed normally more fatly. It is strange that it was sshd...

9

Re: and storage

Hello, Sheridan, you wrote: S>> On my observations just at  deadly problems with shortage of storage - was  on which often there was OOM Kill S> It from that " !", and then  forget to add in process of loading increase. Speech not about " ", and about that that the Windows, normally, do not try to kill the system processes at shortage of storage. And in a case  generally, and this machine in particular at OOM Kill it is possible to reboot at once the machine - to try to waste time on recovery senselessly.

10

Re: and storage

Hello, Vanja Pervachev, you wrote: > I will outline also > why  rises  when storage comes to an end? Because starts to drop in a swap and from for it jumps up iowait. All stupidly rests against a disk that outside looks as logs. > at Windows of the such did not watch > better to hang up than  the last process? If you want that it shot processes - disconnect . Well also track as though storage consumption. Lift monitoring (prometej-telegaf-nodaeksporter-grafana ) and look that . The general-purpose council I can not give as a situation strongly different.  it is necessary big ,   and so on. Also do not forget to clarify before iron/vps purchase - and will be how much expensive to add still storages. Because loading can grow SUDDENLY (the startup became popular and millions  came). Well and in an ideal it is necessary to be able to tear capacities in pair commands for the minimum time. Smoke IaC and ansible.

11

Re: and storage

Hello, Somescout, you wrote: S> Speech not about " ", and about that that the Windows, normally, do not try to kill the system processes at shortage of storage. And in a case  generally, and this machine in particular at OOM Kill it is possible to reboot at once the machine - to try to waste time on recovery senselessly. http://rsdn.org/forum/flame.comp/7121660.1 the author: Sheridan Date: 21.04 14:17

12

Re: and storage

Hello, andrey.desman, you wrote: AD> the Swap to include, oom_score_adj to the necessary processes to expose. Normally system init it supports (at least upstart and systemd). Not, well pens, of course, it is possible to correct much, but it all the same clumsy behavior. AD> well normally process also is killed more fatly. It is strange that it was sshd... I have not been assured that it sshd - it was impossible to come neither on ssh, nor from the console .

13

Re: and storage

Hello, Sheridan, you wrote: S> Well also track as though storage consumption. Lift monitoring (prometej-telegaf-nodaeksporter-grafana ) and look that . The general-purpose council I can not give as a situation strongly different.  it is necessary big ,   and so on. S> Also do not forget to clarify before iron/vps purchase - and will be how much expensive to add still storages. Because loading can grow SUDDENLY (the startup became popular and millions  came). S> Well and in an ideal it is necessary to be able to tear capacities in pair commands for the minimum time. Smoke IaC and ansible. At me a desktop 24 +11

14

Re: and storage

Hello, Sheridan, you wrote: S> http://rsdn.org/forum/flame.comp/7121660.1 the Author: Sheridan Date: 21.04 14:17 Once again - in this case speech about behavior  at shortage of storage. It is clear that after adding of storage and swap magnification all became more or less, but here about shortage -  is bright laid down up pads. By the way Windows, even at the full exhaustion of a swap (when curve application ate all storage) this application (more precisely it died when could not select storage any more) simply laid down and continued to work, let and with a scratch of disks.

15

Re: and storage

Hello, Vanja Pervachev, you wrote: > at me desktop > 24 +11  Well here simply track storage. Be ready to logs when you launch  the heavy. Make svap-undressed on each screw and connect with equal precedence (pri=10 in fstab ) Read still here it I  on a desktop at itself I track storage (as a hobby even the Author blinded  : Sheridan Date: 09.04 01:49) and before start for example  I look how many at me still freely. If it is small - I close that that now is not necessary.

16

Re: and storage

Hello, Somescout, you wrote: S>> http://rsdn.org/forum/flame.comp/7121660.1 the Author: Sheridan Date: 21.04 14:17 S> Once again - in this case speech about behavior  at shortage of storage. It is clear that after adding of storage and swap magnification all became more or less, but here about shortage -  is bright laid down up pads. It does not lay down up pads, and drops in a swap that that counts while unnecessary. Estimate  what is the time at you occupies disk writing  gigabyte.

17

Re: and storage

Hello, Sheridan, you wrote: S>> Once again - in this case speech about behavior  at shortage of storage. It is clear that after adding of storage and swap magnification all became more or less, but here about shortage -  is bright laid down up pads. S> it does not lay down up pads, and drops in a swap that that counts while unnecessary. Estimate  what is the time at you occupies disk writing  gigabyte.  and once again: speech about behavior  at a storage lack when it is launched  OOM Kill. What swap in this situation?

18

Re: and storage

Hello, Sheridan, you wrote: S> Hello, Somescout, you wrote: S>>> http://rsdn.org/forum/flame.comp/7121660.1 the Author: Sheridan Date: 21.04 14:17 S>> Once again - in this case speech about behavior  at shortage of storage. It is clear that after adding of storage and swap magnification all became more or less, but here about shortage -  is bright laid down up pads. S> it does not lay down up pads, and drops in a swap that that counts while unnecessary. Estimate  what is the time at you occupies disk writing  gigabyte.  itself drops? The program and all system hangs not

19

Re: and storage

Hello, Vanja Pervachev, you wrote: >  itself drops? > the program and all system All hangs not is true. A disk - a resource not programs, and systems as a whole. And it is occupied.

20

Re: and storage

Hello, Somescout, you wrote: S> Eshyo and once again: speech about behavior  at a storage lack when it is launched  OOM Kill. What swap in this situation? Then tell more in detail what exactly you do and that it when shoots. And that suddenly appears that tail/dev/zero

21

Re: and storage

Hello, Sheridan, you wrote: S> I  on a desktop at myself track storage (as a hobby even the Author blinded  : Sheridan Date: 09.04 01:49) and before start for example  I look how many at me still freely. If it is small - I close that that now is not necessary. In Linux also on-default all so with a swap is that as a desktop at a swap it is useless. However, and as the server not so. Practically it turns out that swap presence is an emergency store that something to manage to draw out nevertheless at shortage of storage. Yes, it is possible thoughtfully  limits and priorities for OOM, to test it in different situations, still , in practice always to buy in addition storages to the demanded is better.

22

Re: and storage

Hello, Vanja Pervachev, you wrote: > I will outline also > why  rises  when storage comes to an end? Because tries to rescue system in the conditions of shortage of resources intensively forcing out the data from storage on a disk, and a disk the device very slow. Depending on adjustments it can be as the smooth almost imperceptible degradation, and "extremal evacuation". > at Windows of the such did not watch > better to hang up than  the last process? Such possibility is available, called OOMKiller (see oom_score_adj). Swapping parameters as are adjusted (see vm.swappiness and adjacent).  on Yandex. A disk

23

Re: and storage

Hello, Anton Batenev, you wrote: AB> Hello, Vanja Pervachev, you wrote: >> I will outline also >> why  rises  when storage comes to an end? AB> because tries to rescue system in the conditions of shortage of resources intensively forcing out the data from storage on a disk, and a disk the device very slow. Depending on adjustments it can be as the smooth almost imperceptible degradation, and "extremal evacuation". Rescues system by hangup? From such rescue an output only

24

Re: and storage

Hello, Sheridan, you wrote: S> And that suddenly appears that tail/dev/zero   And why it should put system?

25

Re: and storage

Hello, Ops, you wrote: S>> And that suddenly appears that tail/dev/zero   Ops> And why it should put system? One of old methods to gobble up storage. Probably already , I , was not interested.