Chunk does not exist error from hypertable_size query

mkbaker · December 2, 2024, 7:37pm

We periodically read the size of our hypertable using “SELECT hypertable_size(‘tablename’)”

Recently we ran into an error during the size query:

Npgsql.PostgresException (0x80004005): 42P01: relation "timescaledb_internal.hyper_117_27588_chunk" does not exist

We’re not sure if this is a problem internal to timescaledb or npgsql but it does seem like an internal problem. Has anyone seen this before? Is there a way to avoid or work around it? Knowing the table size is important, but recording data is more important to us.

jonatasdp · December 2, 2024, 8:19pm

Hi @mkbaker , have you seen if you have any background job that removed this one specifically?

Is it happening all the time?

Can you share what is the chunk time interval and the settings you have in the hypertable?

mkbaker · December 3, 2024, 1:12am

It doesn’t happen all the time. The process had been running without issue for a couple weeks. There is no background task that would target a specific chunk. We do have a retention policy, but I’m not sure when it kicks in or if there is any evidence that it is doing anything in our logging. Chunk time interval is one day. We have enabled compression (30 day interval), other than that not many settings.

Thanks

jonatasdp · December 5, 2024, 1:16am

It looks like something is cached and need to be improved internally in the functions.

github.com

timescale/timescaledb/blob/f1d201e5eed5ab3d1f65c19b4cf3cc9a6ea9969e/sql/size_utils.sql#L78


      
                  COALESCE((relsize).toast_size, 0) AS toast_bytes,
                  0::BIGINT AS compressed_total_size,
                  0::BIGINT AS compressed_index_size,
                  0::BIGINT AS compressed_toast_size,
                  0::BIGINT AS compressed_heap_size
              FROM
                  _timescaledb_catalog.hypertable ht
                  JOIN pg_class c ON relname = ht.table_name AND c.relkind = 'r'
                  JOIN pg_namespace n ON n.oid = c.relnamespace
                  AND n.nspname = ht.schema_name
                  JOIN LATERAL _timescaledb_functions.relation_size(c.oid) AS relsize ON TRUE
              WHERE
                  schema_name = schema_name_in
                  AND table_name = table_name_in
          ),
          /* calculate the size of the hypertable chunks */
          _chunk_sizes AS (
              SELECT
                  chunk_id,
                  COALESCE(ch.total_bytes, 0) AS total_bytes,
                  COALESCE(ch.heap_bytes, 0) AS heap_bytes,

I’d recommend you to open a bug on the platform if you come up with any possible reproducible step.

@fabriziomello any ideas?