Refactor autoprewarm_database_main() in preparation for read stream Autoprewarm prewarms blocks from a dump file representing the contents of shared buffers at the time it was dumped. It uses a sorted array of BlockInfoRecords, each representing a block from one of the cluster's databases and tables.
autoprewarm_database_main() prewarms all the blocks from a single database. It is optimized to ensure we don't try to open the same relation or fork over and over again if it has been dropped or is invalid. The main loop handled this by carefully setting various local variables to sentinel values when a run of blocks should be skipped. This method won't work with the read stream API. The read stream callback must be able to advance the current position in the BlockInfoRecord array to allow for reading ahead additional blocks, however a read stream maps 1-1 with a relation and fork combination. So, the main loop in autoprewarm_database_main() must also advance the position in the array of BlockInfoRecords to skip invalid relations and forks. This split control doesn't fit well with the current flow control in autoprewarm_database_main() To make it compatible with the read stream API, change autoprewarm_database_main() to explicitly fast-forward in the BlockInfoRecords array past the blocks belonging to an invalid relation or fork. This commit only implements the new control flow -- it does not use the read stream API. Co-authored-by: Nazir Bilal Yavuz <byavu...@gmail.com> Co-authored-by: Melanie Plageman <melanieplage...@gmail.com> Reviewed-by: Heikki Linnakangas <hlinn...@iki.fi> Reviewed-by: Daniel Gustafsson <dan...@yesql.se> Discussion: https://postgr.es/m/flat/CAN55FZ3n8Gd%2BhajbL%3D5UkGzu_aHGRqnn%2BxktXq2fuds%3D1AOR6Q%40mail.gmail.com Branch ------ master Details ------- https://git.postgresql.org/pg/commitdiff/6acab8bdbcda735ef47b1bb0ba2284d6c465cd88 Modified Files -------------- contrib/pg_prewarm/autoprewarm.c | 172 +++++++++++++++++++++------------------ 1 file changed, 94 insertions(+), 78 deletions(-)
