java -jar ~/sstable-tools-3.9.0-alpha9.jar cqlsh always return Tombstone count of 0.

[jwiencek3]

java -jar ~/sstable-tools-3.9.0-alpha9.jar cqlsh

cqlsh> describe sstables

/data/data/ionic_na/logs-a776ecc131b511e7bfae1bf589cdc0b2/mc-22945-big-Data.db ============================================================================== Partitions: 84792 Rows: 84792 Tombstones: 0 Cells: 1086408

Tombstones returns a value =0

When I run an sstabledump -d |grep 'deletedAt"

I get a very large number of rows returned.

[tolbertam]

Hi @jwiencek3, i'm working to reproduce this. I just did a quick sanity check to see if it's detecting tombstones at all and was able to:

in cqlsh:

create table and create row-level and cell-level tombstones:

cqlsh> create KEYSPACE simple WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1};
cqlsh> use simple;
cqlsh:simple> create table tbl (k int primary key, v int);
cqlsh:simple> create table tbl (k int primary key, v int);
cqlsh:simple> insert into tbl (k, v) values (0, 1);
cqlsh:simple> delete from tbl where k = 0 ;
cqlsh:simple> insert into tbl (k, v) values (1, 0);
cqlsh:simple> delete v from tbl where k = 1;

/.../node1/data0/simple/tbl-2d00e5516c0111e79e27ddcd9a2c9d1b/mc-1-big-Data.db
====================================================================================================
Partitions: 2                                                                   
Rows: 1
Tombstones: 2
Cells: 1
Widest Partitions:
   [1] 1
   [0] 0
Largest Partitions:
   [1] 37 37 B
   [0] 0 0 B
Tombstone Leaders:
   [1] 1
   [0] 1

sstabledump output:

[1]@0 Row[info=[ts=1500413826297213] ]:  | [v=<tombstone> ts=1500413838802882 ldt=1500413838]
[0]@32 deletedAt=1500413809480066, localDeletion=1500413809

Although it's entirely possible there's a case i'm missing. Will think about that.

[tolbertam]

Ah, i had an idea, maybe we aren't checking for Row level deletions. In my test case above i'm doing cell and partition level deletion, but nothing at row level.

[tolbertam]

That seems to work as well:

cqlsh> use simple;
cqlsh:simple> create table tbl2 (k0 int, k1 int, v int, primary key (k0, k1));
cqlsh:simple> insert into tbl2 (k0, k1, v) values (0, 0, 0);
cqlsh:simple> delete from tbl2 where k0=0 and k1=0;

/.../node1/data0/simple/tbl2-4fa2f1f16c0311e79e27ddcd9a2c9d1b/mc-1-big-Data.db
=====================================================================================================
Partitions: 1                                                                   
Rows: 1
Tombstones: 1
Cells: 0
Widest Partitions:
   [0] 1
Largest Partitions:
   [0] 25 25 B
Tombstone Leaders:
   [0] 1

[0]@0 deletedAt=1500414744019701, localDeletion=1500414744
[0]@0 Row[info=[ts=-9223372036854775808] del=deletedAt=1500414751075742, localDeletion=1500414751 ]: 0 | 

I'll keep experimenting but any further info about what level the deletes are being done at would be helpful.

[tolbertam]

I believe I've reproduced the issue. It happens when you have a compound primary key and the partition key is also composed of several columns, i.e.:

cqlsh> use simple;
cqlsh:simple> create table tbl5 (k0 int, k1 int, k2 int, v int, primary key ((k0, k1), k2));
cqlsh:simple> insert into tbl5 (k0, k1, k2, v) values (0, 0, 0, 0);
cqlsh:simple> delete from tbl5 where k0=0 and k1=0 and k2=0;

/.../node1/data0/simple/tbl5-da5725316c0511e79e27ddcd9a2c9d1b/mc-1-big-Data.db
=====================================================================================================
Partitions: 1                                                                   
Rows: 1
Tombstones: 0
Cells: 0
Widest Partitions:
   [0:0] 1
Largest Partitions:
   [0:0] 25 25 B

I'll see if I can fix this quickly.

[tolbertam]

Think i have a fix in #60. If that looks right let me know and i'll make a release, thanks!

[jwiencek3]

I'll test it this morning and get back to you as soon as I can.

Regards

John

[jwiencek3]

I tested the fix and it works as expected. Tombstone count matches total when I do a sstabledump -s |grep "deletedAt |wc -l

[tolbertam]

Awesome, thanks for verifying! I'll do a release tonight 👍