Thanks for confirming "the tombstones will only get removed during compaction
if they are older than GC_Grace_Seconds for that CF". I didn't find such a
clarification in the documentation. That answered my question.
Since the table that has too many tombstones is in the system keyspace, I
cannot alter its gc_grace_seconds setting. gc_grace_seconds is now 7 days,
which is certainly longer than the age of the tombstones.
Is there any way that I can remove the tombstones in the system keyspace
immediately?
At 2015-01-13 19:49:47, "Rahul Neelakantan" <ra...@rahul.be> wrote:
I am not sure about the tombstone_failure_threshold, but the tombstones will
only get removed during compaction if they are older than GC_Grace_Seconds for
that CF. How old are these tombstones?
Rahul
On Jan 12, 2015, at 11:27 PM, Xu Zhongxing <xu_zhong_x...@163.com> wrote:
Hi,
When I connect to C* with driver, I found some warnings in the log (I increased
tombstone_failure_threshold to 150000 to see the warning)
WARN [ReadStage:5] 2015-01-13 12:21:14,595 SliceQueryFilter.java (line 225)
Read 34188 live and 104186 tombstoned cells in system.schema_columns (see
tombstone_warn_threshold). 2147483387 columns was requested, slices=[-],
delInfo={deletedAt=-9223372036854775808, localDeletion=2147483647}
WARN [ReadStage:5] 2015-01-13 12:21:15,562 SliceQueryFilter.java (line 225)
Read 34209 live and 104247 tombstoned cells in system.schema_columns (see
tombstone_warn_threshold). 2147449199 columns was requested, slices=[-],
delInfo={deletedAt=-9223372036854775808, localDeletion=2147483647}
I run the command:
nodetool compact system
But the tombstone number does not decrease. I still see the warnings with the
exact number of tombstones.
Why is this happening? What should I do to remove the tombstones in the system
keyspace?
