Add support to timestamp resolution other than microsecond for TWCS

[raphaelsc]

Certain/older clients may decide to work with milliseconds as timestamp resolution instead

[slivne]

@raphaelsc can you please clarify the issue with different clients ?

if we have two clients writing data with different resolution how will the timestamp info be persisted into the file ?

are other compaction strategies resilient to this - and if so how ?

[raphaelsc]

On Sun, Jan 28, 2018 at 9:20 AM, Shlomi Livne ***@***.***> wrote: @raphaelsc <https://github.com/raphaelsc> can you please clarify the issue with different clients ? if we have two clients writing data with different resolution how will the timestamp info be persisted into the file ?

AFAICT, user supplied timestamp will be persisted as is, so user is responsible to keeping timestamp unit consistent for a given column family. Perhaps we check that a timestamp has millisecond resolution and convert it into micro before forwarding to sstable write? I don't think so. I need to double check this.

are other compaction strategies resilient to this - and if so how ?

I don't think either TWCS or DTCS will be resilient to this. They have the timestamp resolution option, defined in schema, and expects timestamp in sstable metadata (min, max) to be exactly of the unit specified in schema. from datastax doc: "When setting the timestamp_resolution for Date Tiered Compaction Strategy (DTCS) settings, it is important to understand why writes also need to set the same timestamp resolution. Failure to do so can lead to a unmanageable number of sstables which can in turn lead to compaction and repair problems." so same requirement for C*. What we need to do here is to allow an user willing to work with only milliseconds to be able to do so, by extending TWCS to support it. Currently, TWCS will only work with user supplied or server generated timestamp which uses microsecond as resolution.

…

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3152 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABWAc_tEtipvQ1JxyOBYAcOIY3kfYW07ks5tPFfygaJpZM4RseEr> .

[slivne]

@raphaelsc well it seems strange that the user provided timestamp is not validated against the schema - I would expect some validation to be done.

[raphaelsc]

On Sun, Feb 4, 2018 at 10:08 AM, Shlomi Livne ***@***.***> wrote: @raphaelsc <https://github.com/raphaelsc> well it seems strange that the user provided timestamp is not validated against the schema - I would expect some validation to be done.

It's probably a bad design issue. The timestamp resolution option exists only for time series strategies, so it'd have to be done at strategy level. So we'd need to postpone validation to strategy when extracting timestamp from sstables. We could try to add some resilience by converting ts on the fly, or throw exception which would result in shutdown. Either way is better than current because strategy will silently work with whatever timestamp it's provided with which can lead to it misbehaving and consequently user will have bad read performance.

…

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3152 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABWAc_ZwFkJYdIksZdNQjIzzbx5pda-Yks5tRZ2xgaJpZM4RseEr> .

[tzach]

Certain/older clients may decide to work with milliseconds as timestamp resolution instead

which client/versions?
In particular, Kairos 1.2.0 is using cassandra-driver-core 3.3.2[1]
Is this driver version problematic?

[1] https://github.com/kairosdb/kairosdb/blob/v1.2.0/ivy.xml#L61

[gcmcnutt]

we too are blocked on using TWCS with kairosDB -- if anyone can come up with a workaround or hackaround, even in kairosdb, in the mean time, that would help

[avikivity]

Is this about the compaction_window_unit of TWCS? If so, I don't think it affects timestamp resolution.

If not, what is it?

[raphaelsc]

On Sun, Apr 8, 2018 at 8:58 AM, Avi Kivity ***@***.***> wrote: Is this about the compaction_window_unit of TWCS? If so, I don't think it affects timestamp resolution. If not, what is it?

the timestamp resolution provided by client, which goes in timestamp_resolution option which in turn goes in compaction strategy options. That's somewhat awful. Older clients use millisecond instead of microsecond.

…

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#3152 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABWAcwUNLqW2o70K7k9SbAE2Zls8pF7yks5tmft5gaJpZM4RseEr> .

[raphaelsc]

Patch sent, waiting for review...

[avikivity]

Wow, that's truly terrible. Oh well.

[dorlaor]

It will be nice to push to 2.2

…

On Thu, Apr 12, 2018 at 3:35 AM, scylladb-promoter ***@***.*** > wrote: Closed #3152 <#3152> via 0c72781 <0c72781> . — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3152 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABp6RejQTnMSFBPiykh4VIMEdRPMtRq6ks5tny3cgaJpZM4RseEr> .

[tzach]

Not yet part of 2.2
@avikivity can we backport it to 2.2.x?