[crac] RFR: PID adjustment on checkpoint [v16]

Anton Kozlov akozlov at openjdk.org
Fri Jun 30 13:25:39 UTC 2023


On Fri, 30 Jun 2023 13:20:58 GMT, Roman Marchenko <rmarchenko at openjdk.org> wrote:

>> On restore, there might be PID value conflicts because of small PID values, if it was checkpoint'ed in a container. Therefore, when checkpointing in a container, we need to move PID value for new processes to a particular value to avoid conflicts on restore.
>> 
>> See https://github.com/CRaC/example-lambda/blob/master/checkpoint.cmd.sh#L8 for example.
>> 
>> This PR contains implemented functionality similar to the example above, making this work out of the box. By default, if checkpointing, PID is adjusted only if Java's PID is 1 that means Java is run in a container. To adjust PID manually for a checkpoint'ed process, `-XX:CRaCMinPid=<value>` option should be used along with `CRaCCheckpointTo`. Min `CRaCMinPid` value is 1, max `CRaCMinPid` value is `UINT_MAX`, but it is actually limited by OS's pid_max.
>> 
>> There are the following possible scenarios for CRaC running in a container:
>> 
>>     // getpid   CRaCMinPid  |   set_last_pid      fork
>>     // ------------------------------------------------
>>     //   1         -        |    yes (default)    yes
>>     //   1         1        |    no               yes
>>     //   1        >1        |    yes              yes
>>     //   >1        -        |    no               no
>>     //   >1      <=getpid   |    no               no
>>     //   >1       getpid<   |    yes              yes
>
> Roman Marchenko has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Waiting for the last child while spinning PID

Thanks, LGTM!

-------------

Marked as reviewed by akozlov (Lead).

PR Review: https://git.openjdk.org/crac/pull/86#pullrequestreview-1507188765


More information about the crac-dev mailing list