Fork me on GitHub
#onyx
<
2017-05-21
>
michaeldrogalis01:05:35

@lmergen Thanks! We’ll get it looked at asap - everyone in Seattle went outside all day. First day of sunshine in ages. 😛

lucasbradstreet06:05:04

@devth fixed and backported to 0.9.15.1

devth14:05:32

lucasbradstreet: thanks! confirmed it works.

lucasbradstreet06:05:12

I am just stripping out the transaction functions for now.

lmergen06:05:27

@lucasbradstreet great, i see you fixed up my PR 🙂

lmergen06:05:00

for curiosity’s sake, where exactly is where the backpressure comes into play ?

lmergen06:05:48

as far as i can see, you basically have gotten rid of the separation between partitioning and the read step

lucasbradstreet06:05:21

In 0.10, a checkpoint barrier is pushed through the job, with each barrier initiating a checkpoint in each task

lucasbradstreet06:05:35

But you can emit quite a lot of the partitioning segments in that time

lmergen06:05:09

yes, i noticed that was happening — i emitted all the partitions, and the checkpointing didn’t “work”

lucasbradstreet06:05:12

Because a small amount of partitioning data will not fill a small buffer. But then these segments are exploded

lucasbradstreet06:05:40

By the downstream task. So not very good backpressure or checkpointing

lucasbradstreet06:05:52

So it was better to figure out how to partition it in the input task

lucasbradstreet06:05:11

Which I can if all peers agree on the min and max id

lmergen06:05:13

yes, i didn’t want to break backwards compatibility, but this is much better

lucasbradstreet06:05:43

Understood. Yeah, we kinda have to break in this case. Glad it'll be improved anyway

lucasbradstreet07:05:26

Thanks for the PR

lmergen07:05:46

i’ll be trying to get this to work with postgresql today — i noticed some oddities in passing the db configuration, i think the user gets lost somehow i’ll probably add a few more commits to this to fix that as well

lucasbradstreet07:05:34

Great. Would be happy to merge any more work that you can do on it

lellis18:05:28

Hi all! I got this error, but cant understand what can be, some tip?

ERROR [onyx.peer.peer-group-manager:235] - Error caught in PeerGroupManager loop. java.lang.AssertionError: Assert failed: (invariants/short-identifiers-correct %)

lucasbradstreet18:05:14

@lellis hmm. What version of onyx are you seeing this on?

lucasbradstreet18:05:21

We fixed an issue like this a while back

lucasbradstreet18:05:18

If you're on the latest version it should also print out the replica. It'd help a lot if you could give us the full stack trace and replica.

michaeldrogalis20:05:35

@lmergen @lucasbradstreet Really glad that we ditched the partition-keys task in onyx-sql as a two step process to do reading. That was always awkward and a waste of resources.