Hi Brian,

Thanks for the response.

I've managed to pin it down a bit more precisely.

An SQL query by "uuid" in the Phoenix sqlline.py script (JDBC):

0: jdbc:phoenix:10.124.12.191> SELECT * FROM ANALYTICS WHERE "uuid" = 'c21bcfaf-2a2f-4776-9de6-7e25608baf06';
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
|               school_code                |                   code                   |        iso_8601         |                   uuid                   |
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
| WRA0004                                  | site.view                                | 2016-02-18 16:00:00.000 | c21bcfaf-2a2f-4776-9de6-7e25608baf06     |
| WRA0004                                  | site.view                                | 2016-02-18 16:00:00.000 | c21bcfaf-2a2f-4776-9de6-7e25608baf06     |
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+


And a HBASE scan for the RowKey:

base(main):001:0> get 'ANALYTICS', "WRA0004\x00site.view\x00\x80\x00\x01R\xF5\x1C\x04\x00c21bcfaf-2a2f-4776-9de6-7e25608baf06"
COLUMN                                                CELL
 events:_0                                            timestamp=1477549465155, value=
 events:auth.name                                     timestamp=1477549465155, value=NURUL296389@Yes.My
 events:auth.role                                     timestamp=1477549465155, value=User
 events:auth.uuid                                     timestamp=1477549465155, value=23B1BEAB200055F315DE1F303EA4C00141F9704C0A51A527
 events:date                                          timestamp=1477549465155, value=2016-02-19 00:00:00
 events:migrated                                      timestamp=1477549465155, value=\x01
 events:site_uuid                                     timestamp=1477549465155, value=2BAC0082200254FE67D6CF0C5E675D052A3F53BC00260F67
 events:type                                          timestamp=1477549465155, value=site
 events:user_uuid                                     timestamp=1477549465155, value=23B1BEAB200055F315DE1F303EA4C00141F9704C0A51A527

We've been checking, and re-checking - but HBase is adamant it has 745,281 rows, while Phoenix is adamant it has 1,527,352...

Is there any situation where Phoenix and HBase row counts will differ? Some of our records will be similar, but should have different UUIDs. Does Phoenix/HBase de-duplication occur at any point that might cause this?

Thanks for any help you guys can offer,

Craig Roberts
Senior Developer

FrogAsia Sdn Bhd (A YTL Company) Unit 9, Level 2, D6 at Sentul East | 801, Jalan Sentul, 51000 Kuala Lumpur | 01125618093 | Twitter | Facebook | Website

This message (including any attachments) is for the use of the addressee only. It may contain private proprietary or legally privileged statements and information. No confidentiality or privilege is waived or lost by any mistransmission. If you are not the intended recipient, please immediately delete it and all copies of it from your system, destroy any hard copies of it and notify the sender. You must not, directly or indirectly, use, disclose, distribute, print, copy or rely on any part of the message if you are not the intended recipient. Any views expressed in this message (including any attachments) are those of the individual sender and not those of any member of the YTL Group, except where the message states otherwise and the sender is authorized to state them to be the views of any such entity.


On Wed, Oct 26, 2016 at 6:58 PM, Brian Jeltema <bdjeltema@gmail.com> wrote:
I suspect that both rows exist and that they are different,  but you can’t see the difference due to
some non-printing character. Did you try doing the query using JDBC or a direct HBase scan?

Brian

On Oct 26, 2016, at 6:20 AM, Craig Roberts <craig.roberts@frogasia.com> wrote:

Hi all,

I have a strange issue with Phoenix/HBase (Phoenix 4.4 and HBase 1.1 in HDP 2.4).

This is my table DDL:

CREATE TABLE IF NOT EXISTS ANALYTICS(
  "school_code" VARCHAR NOT NULL,
  "code" VARCHAR NOT NULL,
  "iso_8601" DATE NOT NULL,
  "uuid" VARCHAR NOT NULL,
  -- Used to ensure the "events" column family exists
  "events"."___" VARCHAR,
  CONSTRAINT id PRIMARY KEY ("school_code", "code", "iso_8601", "uuid")
)


I then load events into the table using a custom Flume sink.


0: jdbc:phoenix:10.124.12.191> SELECT COUNT(*) FROM ANALYTICS;
+------------------------------------------+
|                 COUNT(1)                 |
+------------------------------------------+
| 1248515                                  |
+------------------------------------------+
1 row selected (2.415 seconds)
0: jdbc:phoenix:10.124.12.191> SELECT COUNT(DISTINCT("uuid")) FROM ANALYTICS;
+------------------------------------------+
|          DISTINCT_COUNT("uuid")          |
+------------------------------------------+
| 773373                                   |
+------------------------------------------+
1 row selected (5.751 seconds)


Now, this is only showing up with lots of events, and I have almost 1 million events.

We've managed to pin down an errant event UUID. The primary key is meant to ensure that repeated events (Flume's at-least-once guarantees) are overwritten, rather than duplicated (due to UPSERT). Here's some example queries for the errant UUID:

0: jdbc:phoenix:10.124.12.191> SELECT * FROM ANALYTICS WHERE "uuid" = '00011783-c870-463c-9c01-4e530a57714f';
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
|               school_code                |                   code                   |        iso_8601         |                   uuid                   |                  
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
| WRA0004                                  | site.view                                | 2016-03-21 16:00:00.000 | 00011783-c870-463c-9c01-4e530a57714f     |                                         
| WRA0004                                  | site.view                                | 2016-03-21 16:00:00.000 | 00011783-c870-463c-9c01-4e530a57714f     |                                         
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+


Which shows two events with what should be a duplicate primary key.

If I pin it down by including all the primary key fields, however, I get one result:

0: jdbc:phoenix:10.124.12.191> SELECT * FROM ANALYTICS WHERE "school_code" = 'WRA0004' AND "code" = 'site.view' AND "iso_8601" = TO_DATE('2016-03-21 16:00:00') AND "uuid" = '00011783-c870-463c-9c01-4e530a57714f';
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
|               school_code                |                   code                   |        iso_8601         |                   uuid                   |
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+
| WRA0004                                  | site.view                                | 2016-03-21 16:00:00.000 | 00011783-c870-463c-9c01-4e530a57714f     |
+------------------------------------------+------------------------------------------+-------------------------+------------------------------------------+


Finally, hbase shell is showing me the correct number of records:

hbase(main):001:0> count 'ANALYTICS'
# lots of output
773376 row(s) in 142.3770 seconds


Now, I'll admit, I would have expected this to be "impossible".

Does anybody have any idea how Phoenix can show me two primary keys with a generic query, but only one for a specific query? And how I might fix this?

Thanks in advance,
Craig Roberts
Senior Developer

FrogAsia Sdn Bhd (A YTL Company) Unit 9, Level 2, D6 at Sentul East | 801, Jalan Sentul, 51000 Kuala Lumpur | 01125618093 | Twitter | Facebook | Website

This message (including any attachments) is for the use of the addressee only. It may contain private proprietary or legally privileged statements and information. No confidentiality or privilege is waived or lost by any mistransmission. If you are not the intended recipient, please immediately delete it and all copies of it from your system, destroy any hard copies of it and notify the sender. You must not, directly or indirectly, use, disclose, distribute, print, copy or rely on any part of the message if you are not the intended recipient. Any views expressed in this message (including any attachments) are those of the individual sender and not those of any member of the YTL Group, except where the message states otherwise and the sender is authorized to state them to be the views of any such entity.