limiting the etcd event history

48 views
Skip to first unread message

riche

unread,
May 15, 2019, 2:19:57 PM5/15/19
to CoreOS User
I thought I had seen an option to limit the number of events in etcd's event history, but am not finding it. Did this ever exist?

Thanks,
_Rich

Brandon Philips

unread,
May 17, 2019, 7:04:04 PM5/17/19
to riche, CoreOS User
Hello RIch-


Brandon

--
You received this message because you are subscribed to the Google Groups "CoreOS User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coreos-user/8566db37-14df-4c25-9262-96ebf01cf727%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

riche

unread,
May 17, 2019, 9:27:21 PM5/17/19
to CoreOS User
Thanks Brandon. My question was not clear. I want to permanently limit the size of the history, as in roll events out when the queue is 100 deep, rather than the current 1000.

We have some devs who insist on regularly stuffing 1M blobs into the database, which results in steading increasing memory and disk usage. The minimum value for auto-compaction-retention is one hour, and these requests will add GBs to etcd's memory usage in one hour.

_Rich

Brandon Philips

unread,
May 20, 2019, 12:54:25 PM5/20/19
to riche, CoreOS User, etcd-dev
--auto-compaction-mode=revision --auto-compaction-retention=100 should do that.

LMK if there is an issue with that set of flags. Also, cc'ing in the etcd mailing list.

--
You received this message because you are subscribed to the Google Groups "CoreOS User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.

riche

unread,
May 20, 2019, 1:13:01 PM5/20/19
to CoreOS User
Thanks Brandon, that is what I was seeking. We'll experiment to find combinations of those two options that will keep these blobs from causing etcd to use too much memory.

_Rich


On Monday, May 20, 2019 at 9:54:25 AM UTC-7, bphilips wrote:
--auto-compaction-mode=revision --auto-compaction-retention=100 should do that.

LMK if there is an issue with that set of flags. Also, cc'ing in the etcd mailing list.

On Fri, May 17, 2019 at 6:27 PM riche <richard...@gmail.com> wrote:
Thanks Brandon. My question was not clear. I want to permanently limit the size of the history, as in roll events out when the queue is 100 deep, rather than the current 1000.

We have some devs who insist on regularly stuffing 1M blobs into the database, which results in steading increasing memory and disk usage. The minimum value for auto-compaction-retention is one hour, and these requests will add GBs to etcd's memory usage in one hour.

_Rich

On Friday, May 17, 2019 at 4:04:04 PM UTC-7, bphilips wrote:
Hello RIch-


Brandon

--
You received this message because you are subscribed to the Google Groups "CoreOS User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coreo...@googlegroups.com.

Brandon Philips

unread,
May 20, 2019, 4:51:07 PM5/20/19
to riche, CoreOS User
This is in the v3 data store right?

To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coreos-user/3aea0e2f-639e-4b55-b7da-c4f6325156d6%40googlegroups.com.

riche

unread,
May 22, 2019, 12:23:28 PM5/22/19
to CoreOS User
Yes it is. With --auto-compaction-retention=1000 it does not gobble up memory and cpu under the load that drove v2 to the limit (oom).

_Rich


On Monday, May 20, 2019 at 1:51:07 PM UTC-7, bphilips wrote:
This is in the v3 data store right?

On Mon, May 20, 2019 at 10:13 AM riche <richard...@gmail.com> wrote:
Thanks Brandon, that is what I was seeking. We'll experiment to find combinations of those two options that will keep these blobs from causing etcd to use too much memory.

_Rich

On Monday, May 20, 2019 at 9:54:25 AM UTC-7, bphilips wrote:
--auto-compaction-mode=revision --auto-compaction-retention=100 should do that.

LMK if there is an issue with that set of flags. Also, cc'ing in the etcd mailing list.

On Fri, May 17, 2019 at 6:27 PM riche <richard...@gmail.com> wrote:
Thanks Brandon. My question was not clear. I want to permanently limit the size of the history, as in roll events out when the queue is 100 deep, rather than the current 1000.

We have some devs who insist on regularly stuffing 1M blobs into the database, which results in steading increasing memory and disk usage. The minimum value for auto-compaction-retention is one hour, and these requests will add GBs to etcd's memory usage in one hour.

_Rich

On Friday, May 17, 2019 at 4:04:04 PM UTC-7, bphilips wrote:
Hello RIch-


Brandon

--
You received this message because you are subscribed to the Google Groups "CoreOS User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coreo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coreos-user/f4052722-f2eb-4c35-aa87-f82822417bda%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "CoreOS User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coreo...@googlegroups.com.

Brandon Philips

unread,
May 23, 2019, 6:15:21 PM5/23/19
to riche, CoreOS User
great! glad this resolved it.

To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coreos-user/47852a48-100c-4c28-b40e-3d82bb4aefcc%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages