Kepner, Jeremy - 0553 - ....
Would there be any interest in developing a SQL-layer on top of Accumulo?
I'm part of the Apache Phoenix project and we've built a similar system on
top of HBase. I wanted to see if there'd be interest on your end at working
with us to generalizing our client and provide in a server that would do
Accumulo-specific push down in support of a SQL layer. I suspect there's
enough similarity between HBase and Accumulo that this would be feasible.
I think there is a great desire for something like this on top of Accumulo.
We've actually had an open JIRA for exposing Accumulo through a query
language since Jan '13 - well over a year now.
It hasn't progressed much beyond the "it would be nice" stage, but you can
track it at
However, I'm not sure that all of us understand the level of work that
would be required for this task (at least, I certainly don't). Can you
provide a slightly more detailed plan for what you envision?
On Tue, Apr 29, 2014 at 1:32 AM, James Taylor <[EMAIL PROTECTED]>wrote:
sql on accumulo would not satisfy "a query language on accumulo" in my opinion. There are certain things that just don't make sense in sql with some more flexible schemas.
I'm actually talking about this subject at the accumulo meetup next week: http://www.meetup.com/Accumulo-Users-DC/events/178839102/
Not being a downer on the idea as i think phoenix support would be cool, just saying a query language to support everything that people have done with accumulo is something other than sql.
Donald, looking forward to your talk.
On Tue, Apr 29, 2014 at 8:04 AM, Donald Miner <[EMAIL PROTECTED]>wrote:
Can you explain how the SQL layer to HBase works?
On Apr 29, 2014, at 1:32 AM, James Taylor <[EMAIL PROTECTED]>
Can you explain how the SQL layer to HBase works?
On Mon, Apr 28, 2014 at 10:32:41PM -0700, James Taylor wrote:
I have taken a quick look at phoenix. It's baked into HBase-specific
features pretty hard.
It uses coprocessors to do things like create index entries. This is a
common enough idiom in the HBase community, but not something we've
supported in Accumulo. In general, you do not want an accumulo Iterator or
Constraint generating data for other tables.
However, a more sophisticated Percolator type implementation (
https://github.com/keith-turner/Accismus) could support index generation
and query transactions.
We could probably re-use a lot of it, but it's not going to be as simple as
changing the classes that talk to the database back-end.
On Tue, Apr 29, 2014 at 9:21 AM, Kepner, Jeremy - 0553 - MITLL <
[EMAIL PROTECTED]> wrote:
Thanks for reaching out.
Like Eric said, I'm a little scared because I know that Phoenix is
rather baked into HBase's API. But, that's half the fun in writing some
new code :)
I'd be happy to help evaluate what this would look like - what is
different (both good and bad) in Accumulo. Like was previously
mentioned, targeting the Accismus (Percolator) prototype to generate the
secondary indices would, IMO, be the best target here. I know it's in
the very early stages right now, but I still believe that it would be
the long-term solution.
On 4/29/14, 1:32 AM, James Taylor wrote:
@Mike - thanks for pointing out that JIRA. I'll comment there with more
detail. My high-level thinking would be to work with your community to do a
feasibility study and perhaps POC. I'd be, of course, relying on your
expertise of Accumulo, as my knowledge is pretty limited.
@Jeremy - take a look at the prior presentations to get a better idea:
http://phoenix.incubator.apache.org/resources.html. In particular, take a
look at the ApacheCon presentation and the kinds of pushdown we do to the
@Josh - it's less baked in than you'd think on the client where the query
parsing, compilation, optimization, and orchestration occurs. The
client/server interaction is hidden behind the ConnectionQueryServices
interface, the scanning behind ResultIterator (in
particular ScanningResultIterator), the DML behind MutationState, and
KeyValue interaction behind KeyValueBuilder. Yes, though, it would require
some more abstraction, but probably not too bad, though. On the
server-side, the entry points would all be different and that's where I'd
need your insights for what's possible.
@Donald - you make a good point. We've stretched the capabilities of SQL,
especially around DDL to support views (
http://phoenix.incubator.apache.org/views.html) which allow you to add new
columns, and read-time schema (
http://phoenix.incubator.apache.org/dynamic_columns.html) which allow you
to specify column definitions at read/write time.
@Eric - I agree about having txn support (probably through snapshot
isolation) by controlling the timestamp, and then layering indexing on top
of that. That's where we're headed. But I wouldn't let that stop the effort
- it would just be layered on top of what's already there. FWIW, there's
another interesting indexing model that has been termed "local indexing"(
https://github.com/Huawei-Hadoop/hindex) which is being worked on right now
(should be available in either our 4.1 or 4.2 release). In this model, the
table data and index data are co-located on the same region server through
a kind of "buddy" region mechanism. The advantage is that you take no hit
at write time, as you're writing both the index and table data together.
Not sure how/if this would transfer over to the Accumulo world.
TLDR? Let's continue in the JIRA?
On Tue, Apr 29, 2014 at 7:45 AM, Josh Elser <[EMAIL PROTECTED]> wrote:
Definitely. I'm a little concerned about what's expected to be provided
by the "database" (HBase, Accumulo) as I believe HBase is a little more
flexible in allowing writes internally where Accumulo has thus far said
"you're gonna have a bad time".
Interesting. Given that Accumulo doesn't have a fixed column family
schema, this might make index generation even easier (maybe "cleaner" is
the proper word). You could easily co-locate the indices with the data,
given them a proper name.
Problem still exists that we don't have a solid way to do this solely
inside of Accumulo ATM. I'd imagine that if someone stepped up to
implement coprocessors, we'd be taking the route of a separate,
standalone process (as opposed to in-RegionServer). Hypothetically, we
could do the same for Phoenix in the short-term.
Can you quantify what would be expected by Accumulo to integrate with
Phoenix (maybe list what exactly is done inside of HBase at a high
level?) so that we could give some more targeted ideas/feelings as to
what the level of work would be inside Accumulo?
Mailing list is fine by me for while we get this hashed out :). We can
move to Jira when we start getting into specifics.
On Tue, Apr 29, 2014 at 11:57 AM, Josh Elser <[EMAIL PROTECTED]> wrote:
Tell me more about what you mean by "allowing writes internally".
With HBase, you can do something similar (though, you're right, you'd need
to create the column family upfront or take the hit of creating it
dynamically - that's a nice feature that Accumulo has). The reason this
doesn't work is that you need a different row key so that the index rows
are ordered according to their indexed column values. If you put it in a
column family of the data table, they're ordered in the same way as the
data table. This makes range scans over index tables very expensive, as the
rows would need to be re-ordered.
There's not a lot of hard/fast requirements. Most of what Phoenix does is
to optimize performance by leveraging the capabilities of the server. In
terms of hard/fast requirements, these come to mind:
- data is returned in row key order from range scans
- a scan may set a start key/stop key to do a range scan
- a row key may be composed of arbitrary bytes
- a client may "pre-split" a table by providing the region boundaries at
table create time (we rely on this for salting to prevent hotspotting:
- the client has access to the region boundaries of a table (this allows
for better parallelization)
- the client may issue chunk up a scan into smaller, multiple scans and run
them in parallel
Some of these may be a bit squishy, as there may be existing machinery
already in your client programming model that could be leverage. The client
API of HBase, for example, does not provide the ability out of the box to
parallelize a scan, so this is something Phoenix had to add on top (through
chunking up scans at or within region boundaries).
Phoenix manages the metadata of your tables (tables, columns, indexes,
views, etc) in an HBase table. DDL statements such as CREATE TABLE, DROP
TABLE, ALTER TABLE are atomic, transactional operations b/c we don't want
our metadata table to get in a corrupt state. To accomplish this, we rely
- setting a "split policy" that ensures that the table data for a given
"tenant" (we support multi-tenancy:
http://phoenix.incubator.apache.org/multi-tenancy.html) stay together in
the same region.
- putting the data using an API that guarantees that either the entire
batch of mutations succeed or fail completely.
Again, these are details of our implementation on HBase which do not
necessarily need to be implemented in the same way on a different system.
Phoenix supports sequences which are atomically incremented values. This is
done through a coprocessor currently, due to some limitations with the
HBase Increment API, but the idea is the same as an atomic increment.
Phoenix does the following push down:
- the WHERE clause gets transformed into three things: a start/stop key of
a scan, a skip scan filter to efficiently navigate the key space (see
and a custom filter to rule out a row based on some java code that does
- the GROUP BY clause gets pushed to the server and a coprocessor runs the
scan on each region so that the client doesn't have to get back all the raw
data. Instead, the client gets back the aggregated data (to conserve
network bandwidth and to run the scan where the data lives). The client
then does a final merge sort.
- the ORDER BY clause used in combination with the LIMIT clause is a TopN
query. We optimize this by each region holding on to the top N values with
the client then doing a merge sort with the limit applied.
- the ORDER BY clause on it's own gets executed on each region (spooled
using memory mapped files) and then the client does a merge sort. This
spooling could potentially be done on the client side.
- joins are executed as a broadcast hash join. We run one side of the query
(with the filters applied), compact the results, and send them to each
region server where they are cached while we run the other side of the
query. A coprocessor then does a map lookup (equi-joins only are supported
currently) to join based on the join key and returns the joined results
(i.e. the concatenated values in a single, condensed key value as access
from the client is positional post-join).
For our global secondary indexes (local secondary indexes are different as
we discussed already), we trap updates to the data table through a
coprocessor. For index maintenance you need to know when a change occurs to
a data row what the prior value of the row was. The reason is because you
need to delete the index row corresponding to the old data row and then
insert the index row corresponding to the new value (remember, the index
value makes up the row key). By doing this operation through a coprocessor,
we know that we can get the prior data row state locally. We still need to
issue a Put from one region server to another, but this isn't really an
extra hop, as if it was done on the client, the same hop would need to be
done (but the old row state would need to be pulled over to the client
which is not necessary with the coprocessor based approach). For more on
global secondary indexing, see
http://phoenix.incubator.apache.org/secondary_indexing.html (there are some
good presentations at the end of the page that provide more technical
Phoenix also allows "point-in-time" queries where a client may establish a
connection at an earlier timestamp. If your table is setup to keep multiple
versions of the same row, then you can query "back-in-time" and will see
the data as it was at that point. We more or less get this for free with
the MVCC model of HBase by specifying a max timestamp on a scan. One
slightly tricky bit is we correlate the current DDL of your table based on
the same timestamp as with your data. So when you go back-in-time like
this, you'll also see the structure of your table as it was at time also.
So we do rely on c
On 4/30/14, 3:33 AM, James Taylor wrote:
Haha, sorry, that was a sufficiently ominous statement with insufficient
For discussion sake, let's just say HBase coprocessors and Accumulo
iterators are equivalent, purely in the scope of "running server-side
code" (in the RegionServer/TabletServer). However, there is a notable
difference in the pipeline where each of those are implemented.
Coprocessors have built-in hooks that let you get updates on
PUT/GET/DELETE/etc as well as pre and post each of those operations. In
other words, they provide hooks at a "high database level".
Iterators tend to be much closer to the data itself, only dealing with
streams of data (other iterators stacked on one another). Iterators
implement versioning, visibilities, and can even implement complex
searches. The downside of this approach is that iterators lack any means
to safely write data _outside of the sorted Key-Value pairs in the
tablet currently being processed_. It's possible to make in tablet
updates, but sorted order within a large tablet might make this
difficult as well.
This is why I was thinking percolator would be a better solution, as
it's meant for handling updates like this server-side. However, I
imagine it would be possible, in the short-term, to make some separate
process between Phoenix and Accumulo which handles writes.
Ah, of course. You need the term up front to make it sort properly.
All of these look fine. The Accumulo BatchScanner does that
parallelization for you which is really nice (handling tablet migration
and all that fun stuff transparently).
I'd have to look again at how our mutation failures are handled (or
someone else can chime in). This might be something to keep an eye on
depending on the distribution of mutations in regards to tables.
Conditional Mutations in the about-to-be-released version 1.6.0 will
I've written an iterator to do a group by previously. Depending on the
schema this is fine.
This is an interesting one. If you remove the possibility of tablets
splitting out from underneath you and you had a view of the splits, you
could probably pull it off.
Unless we can do some trickery with the schema, yeah, client side.
The join approach would need to be implemented some other way for the
earlier stated comparison of iterators and coprocessors.
Right, you want to remove the old index value and update a new index
value (actually being two unique keys) in the same transaction to ensure
a valid index. Or, at least ensure that you never remove the old value,
and die before inserting the new value.
Again, not going to work well in an iterator.
I don't see this as a problem. As long as we remove the versioning
iterator from a table (which keeps the most recent version of a key by
default), it should be pretty easy to implement an iterator which
adheres to the "max timestamp" semantics.
Thanks for the explanations, Josh. This sounds very doable. Few more
comments inline below.
On Wed, Apr 30, 2014 at 8:37 AM, Josh Elser <[EMAIL PROTECTED]> wrote:
Another fallback might be to do global index maintenance on the client.
It'd just be more expensive, especially if you want to handle out-of-order
updates (which are particularly tricky, as you have to get multiple
versions of the rows to work out all the different scenarios here).
A second fallback might be to support only local indexing. Does Accumulo
have the concept of a "custom load balancer" that would allow you to
co-locate two regions from different tables? The local-index features has
kind of driven some feature requests on that front for HBase - mainly
callbacks when a region is split or re-located. The rows of the local index
are prefixed with the region start key to keep them together and identify
That's nice that Accumulo has this built-in. Does it allow the client to
specify the split points for the scan in some way?
Client-side could be another fallback. The coprocessor approach is really
only a big win in two cases: if you have a join which doesn't have many
matches (as those rows get filtered on the server-side), or for correlated
sub queries or exists queries where you can filter or collapse many rows to
one or none on the server-side rather than return them all to the client.
The wikisearch example provides something similar to a local index. Rather
than stuff things into two tablets, a single row in accumulo contains both
the index and data stored in separate column families. Iterator trees are
used to execute queries and retrieve data with that row.
On Thu, May 1, 2014 at 2:24 AM, James Taylor <[EMAIL PROTECTED]> wrote:
On 5/1/14, 2:24 AM, James Taylor wrote:
Agreed with what Bill said. Co-locating indices within the same row
simplifies this a bit, IMO.
Assuming I understand properly, you don't need to be cognizant of the
splits. You just specify the Ranges (where each Range is a start key and
end key) and the Accumulo client API does the rest. You can be efficient
by structuring your data so that you don't touch every tabletserver for
every query -- this seems to be what's being suggested.
What do you think is next, James?
I know I won't have a lot of time to devote into heavy development with
what I've already signed up for in the next few months, but I'd still
like to try to help out where possible. Is anyone else on the Accumulo
side interested in getting involved?
Sorry for the delay in getting back to you - things got a bit crazy with
our graduation and HBaseCon happening at the same time.
@Josh & Bill - r.e. Co-locating indices within the same row simplifies this
The secondary indexes need to be in row key order by the indexed columns,
so co-locating them in the data table wouldn't allow the lookup and range
scan abilities we'd need. The advantage of the index is that you don't need
to look at all the data, but can do a point lookup or range scan based on
the usage of the indexed columns in a query.
@Josh - r.e. Assuming I understand properly, you don't need to be cognizant
of the splits. You just specify the Ranges (where each Range is a start key
and end key) and the Accumulo client API does the rest.
Typically the Ranges are merge sorted on the client, so this might require
an extension to the Accumulo client.
r.e. Next steps.
We'd definitely need an expert on the Accumulo side to proceed. I'm happy
to help on the Phoenix side - I'll post a note on our dev list too to see
if there are other folks interested as well. Given the similarities between
Accumulo and HBase and the abstraction Phoenix already has in place, I
don't think the effort would be large to get something up and running.
Maybe a phased approach, would make sense: first with query support and
next with secondary index support?
Not sure where this stacks up in terms of priority for you all. At
Salesforce, we saw a specific need for this with HBase, the "big data
store" on top of which we'd choose to standardize. We realized early on
that we'd never get the adoption we wanted without providing a different,
more familiar programming model: namely SQL. Since we were targeting
supporting interactive web-based applications, anything map/reduce based
wasn't a fit which led us to create Phoenix. Perhaps there are members in
your community in the same boat?
On Fri, May 2, 2014 at 1:44 PM, Josh Elser <[EMAIL PROTECTED]> wrote:
So there may be a bit of confusion with storing index and data in the same
row. By "row" I just mean the logical Accumulo unit, not a "row" as in
"thing in my relational table." Synonyms for "row" in this scheme are
"shard" and "document partition".
You can store multiple documents and indices for those documents in
different column families within the same row. You then have separate
readers for the indices and document data ("sources" in Iterator terms).
Point and range queries are still possible in this fashion, and are made
even easier if there's another level that maps terms to
rows/shards/partition. The wikisearch example is an (admittedly rough)
implementation of this.
I think looking at how "buddy" regions work may help clarify things, since
I imagine it works similarly. If the coprocessor is just reading from a
region "I", that that contains index data for only region "D", then that
maps pretty well to an iterator scanning index data from a column family
"I" and fetching documents from a column family "D".
On Thu, May 8, 2014 at 1:09 AM, James Taylor <[EMAIL PROTECTED]> wrote:
@William - it's entirely possible that my HBase terminology is not mapping
well to Accumulo terminology. If Accumulo has a capability not present in
HBase that'll handle this, that'd be great.
In HBase terminology, by row I mean all of the key values across all column
families with the same row key (Row ID in Accumulo?). So in HBase, it
doesn't work to store the index data in a separate column family for the
same row, because the rows are ordered according to the data table row key.
We need the rows of an index to be ordered by the row key formed by the
indexed columns instead. Otherwise we have to re-sort the rows which is
more expensive than just doing a scan over the data table.
With buddy regions, the two regions are from different tables with
different row key orders. All of the data from "D" for a given region is
contained in the buddy region for "I", but in a different order. We equally
rely on the buddy region for "I" being in row key order according to the
indexed columns (as opposed to the row key order of the data table).
On Sat, May 10, 2014 at 7:21 PM, William Slacum <
[EMAIL PROTECTED]> wrote:
On 5/11/14, 12:22 AM, James Taylor wrote:
(sorry for the delay, still trying to stay on top of mail from the outage)
I think I know what Bill is trying to get at here and it hinges on the
fact that Accumulo doesn't require you to define the column families for
a table up front (it has a default locality group which all colfams
which don't have a locality group defined go into -- differs from HBase
where locality group == colfam).
Because of this, you can use the column family and qualifier to get the
properly sorting index records instead of using the row key (assuming
the row is just some bucket/partitioning element). Thus, you can
co-locate index and data key-values within the same row if you're tricky
enough with how you create the table. :)
That sounds promising, Josh & William. Is there a performance penalty with
this approach (versus traversing the rows in row key order)?
On Fri, May 16, 2014 at 8:27 AM, Josh Elser <[EMAIL PROTECTED]> wrote: