Clean out priority blocking stuff in PersistenceThread

Conflicts:
	storage/src/tests/persistence/filestorage/filestormanagertest.cpp

Resolve merge conflict.

1 files changed, 4 insertions, 390 deletions

diff --git a/configdefinitions/src/vespa/stor-filestor.def b/configdefinitions/src/vespa/stor-filestor.def
index 80994bfb477..d20462c99ad 100644
--- a/configdefinitions/src/vespa/stor-filestor.def
+++ b/configdefinitions/src/vespa/stor-filestor.def
@@ -1,79 +1,8 @@
 # Copyright 2017 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 namespace=vespa.config.content
 
-## Use the new storage core
-use_new_core bool default=false restart
-
-## FILE LAYOUT PARAMETERS
-
-## Number of directories per level to spread files across
-## DEPRECATED: see stor-memfilepersistence config instead
-dir_spread int default=256 restart
-
-## Number of directory levels
-## DEPRECATED: see stor-memfilepersistence config instead
-dir_levels int default=1 restart
-
-## FILE SIZE PARAMETERS
-
-## Minimum number of meta data entries in one slotfile. When creating new
-## files or resizing files, enforce it to contain at least this many meta
-## entries. Set to 512 by default, using 20544 bytes total for metadata in
-## new files.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_file_meta_slots int default=512
-
-## Maximum number of entries in one slotfile. File must be split before
-## accepting more data
-##
-## Default ensure meta data is less than 512 kB.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_file_meta_slots int default=13106
-
-## Minimum size of header block. At least this amount of header space will
-## be available in new or resized files. For 512 documents, this will be
-## 200 bytes per document.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_file_header_block_size int default=102848
-
-## Maximum header block size (in bytes). Since the whole meta data list and
-## header block needs to be read into memory for some operations, a limit
-## can be set for the header block, to avoid consuming too much memory.
-##
-## Default is set high, as we dont configure it automatically right now, so we
-## would rather people configure it down than up.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_file_header_block_size int default=33554432
-
-## Minimum size of a single slotfile. When creating or resizing files, they
-## will never become smaller than this. Default of 1 MB, will be 1807 byte
-## per doc if we use all 512 meta data entries set as default min meta
-## entries.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_file_size int default=1048576
-
-## Maximum size of a single slotfile. File must be split before accepting
-## more data. Will return file full errors.
-##
-## Default is set high, as we dont configure it automatically right now, so we
-## would rather people configure it down than up.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_file_size int default=268431360
-
-## When creating new files, always create files as a multiplum of this size.
-## Should be set to the block size on the disk, or possibly a multiplum of
-## that, though we know no advantage of setting it to more than the block
-## size. If you want more free space in files, you should rather adjust that
-## with growfactor or min file sizes.
-## DEPRECATED: see stor-memfilepersistence config instead
-file_block_size int default=4096
-
 ## DETECT FAILURE PARAMETERS
 
-## If persistence operations take more than this amount of milliseconds, it
-## will be logged as a warning.
-warn_on_slow_operations int default=5000
-
 ## After seeing given number of errors on a disk, storage will disable the
 ## disk and restart. If set to 0, storage will never disable disks. Note
 ## that if you get disk errors, which arent automatically fixed, this will
@@ -92,60 +21,6 @@ fail_disk_after_error_count int default=1 restart
 ## during normal load.
 disk_operation_timeout int default=0 restart
 
-## Timestamps provided for entries written are normally provided externally.
-## To detect clock skew between nodes, storage will warn if timestamps look
-## out of sync. Future time clearly indicates clock skew and thus the limit
-## here is low. Time in the past might just indicate that the operation has
-## been queued for a while, so the default value here is not very strict.
-## Time is given in seconds. Note that there are valid usecases for
-## inserting data with old timestamps, for instance when synchronizing two
-## copies of the same data. Warnings should not be printed in these cases.
-time_future_limit int default=5
-time_past_limit int default=3600
-
-## Enabling debug verifications will make storage do extra verifications
-## to find errors as soon as possible. These extra verifications will use up
-## a lot of resources though, and should not be needed in normal operations.
-## They are mostly to be used during test phases or when debugging problems.
-## The value itself is a bit mask, where you can enable different kinds of
-## verifications by setting given bits.
-debug_verifications int default=0 restart
-
-## CONSISTENCY PARAMETERS
-
-## If true, fsync after all disk operations, to ensure no dirty OS file
-## cache afterwards. This is only useful if using cached IO, which is not
-## recommended to start with.
-## DEPRECATED: see stor-memfilepersistence config instead
-fsync_after_each_operation bool default=false restart
-
-## Time period to keep all updates (given in seconds). One can revert any
-## operation done within this time.
-## DEPRECATED: see stor-memfilepersistence config instead
-revert_time_period int default=300
-
-## If a remove is older than the reverttimeperiod, the document it is
-## removing may be removed from the file. There are a few advantages to
-## keeping removes or a bit longer though. If you use this copy to
-## synchronize another copy of data, having the remove entry makes it easy
-## to detect that you should delete this entry from the other data copy.
-## This is useful for internal synchronization of files within VDS if you
-## use multiple copies, or for partial recovery or BCP situations. To
-## guarantuee consistency in such situations, a data destination that have
-## been unavailable for longer than this amount of time, should have its
-## data cleared before being set into the system again. This value is given
-## in seconds, with the default being one week
-## DEPRECATED: see stor-memfilepersistence config instead
-keep_remove_time_period int default=604800
-
-## Maximum number of entries to keep for a single document. Heavy resending
-## within the revert time period may add loads of entries to a file, ruining
-## performance. This setting sets a maximum number of entries for each
-## document. This will make entries potentially live shorter than the revert
-## time period to avoid a resending worst case. A value of 0 means infinite.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_versions_of_single_document_stored int default=0
-
 ## PERFORMANCE PARAMETERS
 
 ## Number of threads to use for each mountpoint. VDS needs memory per thread
@@ -156,11 +31,10 @@ maximum_versions_of_single_document_stored int default=0
 ## See benchmarks for performance/memory tradeoff.
 threads[].lowestpri int default=255 restart
 
-## Pause operations (and block new ones from starting) with priority 
-## lower than this value when executing operations with higher pri than 
-## min_priority_to_be_blocking
-max_priority_to_block int default=255 restart
-min_priority_to_be_blocking int default=0 restart
+## When merging, if we find more than this number of documents that exist on all
+## of the same copies, send a separate apply bucket diff with these entries
+## to an optimized merge chain that guarantuees minimum data transfer.
+common_merge_chain_optimalization_minimum_size int default=64 restart
 
 ## Chunksize to use while merging buckets between nodes.
 ##
@@ -169,173 +43,6 @@ min_priority_to_be_blocking int default=0 restart
 ## while still reading 4k blocks from disk.
 bucket_merge_chunk_size int default=4190208 restart
 
-## When reading a slotfile, one does not know the size of the meta data
-## list, so one have to read a static amount of data, and possibly read more
-## if one didnt read enough. This value needs to be at least 64 byte to read
-## the initial header stating the true size of the meta data list.
-## DEPRECATED: see stor-memfilepersistence config instead
-initial_index_read int default=61440
-
-## Similar to index read, when reading the document identifiers, one does
-## not know the length of the name prior to reading. Thus, if you read less
-## than the size, you will have to do an extra read to get the rest. If you
-## use very long document identifiers you should increase this value to be
-## larger than most of your identifiers.
-## restart flag was added automatically and needs to be verified.
-## DEPRECATED: see stor-memfilepersistence config instead
-initial_name_read int default=512 restart
-
-## When we need to read (or write) multiple entries in a file where we can
-## either read a big enough section to cover all of them. But at some
-## distance between the areas we need, it becomes beneficial to do multiple
-## reads rather than to read over them. This setting set how many sequential
-## bytes we dont need that we allow to be read/written in order to join two
-## logical IO operations together in the application. Setting this low might
-## be ok if system calls are cheap and we manage to queue up next IO
-## operation in time such that the disk dont need to spin an extra round.
-## Setting it high will make the disk more likely to process them together,
-## but the time used to read/write the gap might have been used to do
-## something useful instead.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_gap_to_read_through int default=65536
-
-## Currently not in here as we dont have append functionality yet. Might
-## improve performance in some cases.
-## max_file_appends int default=0
-
-## When writing with direct IO, we need to align blocks to 512b, and to
-## avoid reading we write garbage after each doc to fill 512b block. When
-## resizing or splitting the file we can realign the files such that we
-## remove the gaps of existing data, as we will rewrite everything anyhow.
-## If using very small documents this might improve your disk space
-## utilization. For larger documents it doesnt really reduce much, so might 
-## be useful to turn it off to save CPU.
-## DEPRECATED: see stor-memfilepersistence config instead
-remove_file_gaps_on_rewrite bool default=true restart
-
-## The order we try to enforce in file body blocks. Visitors that only need
-## to visit some of the data in a bucket will be able to read less if what
-## it needs to read is located next to each other on disk. However, as we
-## dont enforce order on write operations, this demands that resize/split
-## operations do the resorting, which, if we cant do it all in memory is
-## very expensive. ANY will not do any reordering. Timestamp will enforce
-## timestamp order, which is fairly close to order that will normally be
-## written anyway, so it should be cheap to reorder even if we cant do it
-## all in memory. This might be useful if you often visit subsets based on
-## time. RGID uses reverse GID, which stores data from one location
-## together. This is useful if you want to visit data from one user from
-## buckets that have many users often. This is a much more expensive sort
-## though, and should only be done if we have enough memory.
-## DEPRECATED: see stor-memfilepersistence config instead
-body_block_order enum { ANY, TIMESTAMP, RGID } default=ANY restart
-
-## If set to true, we will refuse to do reordering of memory unless we have
-## enough memory to do it all in memory. See body_block_order comments.
-## DEPRECATED: see stor-memfilepersistence config instead
-only_reorder_body_in_memory bool default=true restart
-
-## Whether or not we should verify checksums of all read data during regular
-## operations like put, get & remove. Note that some operations, like merge
-## and bucket integrity verify operations will still check checksums even if
-## this is set false.
-## DEPRECATED: see stor-memfilepersistence config instead
-verify_checksum_on_regular_load bool default=true restart
-
-## For streaming search, visiting is very performance critical. Thus you can
-## specifically disable checksum verification for visiting.
-## DEPRECATED: see stor-memfilepersistence config instead
-verify_checksum_on_visit bool default=true restart
-
-## Maximum size of index buffer that will be allowed to stay in memory and
-## not being reduced back to this size after we no longer need it.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_sustainable_index_buffer_size int default=1044480 restart
-
-## Maximum size of input buffer that will be allowed to stay in memory and
-## not being reduced back to this size after we no longer need it.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_sustainable_input_buffer_size int default=1044480 restart
-
-## Maximum size of output buffer that will be allowed to stay in memory and
-## not being reduced back to this size after we no longer need it.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_sustainable_output_buffer_size int default=1044480 restart
-
-## Whether to downsize index buffer immediately after usage if its above the
-## maximum size. If not, it will not be resized down until someone requests
-## to use it that needs less data in the buffer. Index buffer is used all
-## the time so there should be little reason for immediately downsizing it.
-## DEPRECATED: see stor-memfilepersistence config instead
-downsize_index_buffer_immediately_after_use bool default=false restart
-
-## Whether to downsize input buffer immediately after usage if its above the
-## maximum size. If not, it will not be resized down until someone requests
-## to use it that needs less data in the buffer. Input buffer is not used
-## that often, so downsizing it immediately might save some memory.
-## DEPRECATED: see stor-memfilepersistence config instead
-downsize_input_buffer_immediately_after_use bool default=true restart
-
-## Whether to downsize output buffer immediately after usage if its above
-## the maximum size. If not, it will not be resized down until someone
-## requests to use it that needs less data in the buffer. Input buffer is
-## not used that often, so downsizing it immediately might save some memory.
-## DEPRECATED: see stor-memfilepersistence config instead
-downsize_output_buffer_immediately_after_use bool default=true restart
-
-## Minimum size of buffer used to write a continuous file. If maximum amount
-## of memory is not available. At least this amount will be allocated.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_continuous_file_write_buffer_size int default=1044480 restart
-
-## Maximum size of buffer used to write a continuous file. If set above max
-## file size we will always write new files in one go, which probably makes
-## for the least chance of getting fragmentation, but will also consume the
-## most memory. Default of writing a MB at a time, should make for little
-## performance loss because of disk seek time, and hopefully get little
-## fragmentation while keeping memory usage down.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_continuous_file_write_buffer_size int default=1044480 restart
-
-## Minimum amount of memory allocated to read source data during join. This
-## amount of memory will be forced allocated.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_join_source_body_read_buffer_size int default=1044480 restart
-
-## This sets the maximum size of the buffer used in join to read source
-## data. Join uses the least IO if this buffer is as big as the body block
-## of the source file. Due to the memory manager, each join might get that
-## much memory though.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_join_source_body_read_buffer_size int default=16773120 restart
-
-## Minimum amount of memory allocated to read source data during export. This
-## amount of memory will be forced allocated.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_export_source_body_read_buffer_size int default=1044480 restart
-
-## This sets the maximum size of the buffer used in export to read source
-## data. Export uses the least IO if this buffer is as big as the body block
-## of the source file. In addition, reordering of body block might not be
-## feasibly unless the buffer is big enough to include the whole body block.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_export_source_body_read_buffer_size int default=33550336 restart
-
-## Minimum size of buffer used to read data during defragmentation.
-## DEPRECATED: see stor-memfilepersistence config instead
-minimum_defrag_source_body_read_buffer_size int default=1044480 restart
-
-## This sets the maximum size of the buffer used in defragmentation to read
-## source data. Defragmentation uses the least IO if this buffer is as big
-## as the body block of the source file, but this might consume some memory.
-## Defragmentation is not enabled by default.
-## DEPRECATED: see stor-memfilepersistence config instead
-maximum_defrag_source_body_read_buffer_size int default=1044480 restart
-
-## When merging, if we find more than this number of documents that exist on all
-## of the same copies, send a separate apply bucket diff with these entries
-## to an optimized merge chain that guarantuees minimum data transfer.
-common_merge_chain_optimalization_minimum_size int default=64 restart
-
 ## When merging, it is possible to send more metadata than needed in order to
 ## let local nodes in merge decide which entries fits best to add this time
 ## based on disk location. Toggle this option on to use it. Note that memory
@@ -344,100 +51,7 @@ common_merge_chain_optimalization_minimum_size int default=64 restart
 ## fill all.
 enable_merge_local_node_choose_docs_optimalization bool default=true restart
 
-## If set, when we need to cache the entire header, and we already have cached
-## all the metadata, read the metadata to find max header position, and only
-## read the part containing information.
-## DEPRECATED: see stor-memfilepersistence config instead
-read_only_used_header_part_when_possible bool default=true restart
-
-## Enable the slotfile read cache. The cache holds recent metadata and header
-## blocks read from disks, and even if small, is very useful for localized
-## access.
-## DEPRECATED: see stor-memfilepersistence config instead
-enable_slotfile_cache bool default=true restart
-
-## Let the slotfile cache the whole header on a header only operation not
-## needing the entire header, if this amount of header only accesses needing
-## part of the header has already happened.
-##
-## Set very high to begin with to never reduce performance. If you have heavy
-## header only access to some files, you may get better performance by tuning
-## this value.
-## DEPRECATED: see stor-memfilepersistence config instead
-slotfile_precache_header_after_access_count int default=512 restart
-
 ## Whether or not to enable the multibit split optimalization. This is useful
 ## if splitting is expensive, but listing document identifiers is fairly cheap.
 ## This is true for memfile persistence layer, but not for vespa search.
 enable_multibit_split_optimalization bool default=true restart
-
-## STORAGE SPACE vs IO/CPU PERFORMANCE OPTIONS
-
-## If true, use direct IO, bypassing OS caches for disk access. This is very
-## useful as VDS does a random distribution dividing load, so it is unlikely
-## that the OS cache will ever hit, and thus it is a huge performance drain.
-## DEPRECATED: see stor-memfilepersistence config instead
-use_direct_io bool default=true restart
-
-## All IO operations will be aligned to this amount of bytes if direct IO is
-## enabled.
-## DEPRECATED: see stor-memfilepersistence config instead
-block_alignment_size int default=512 restart
-
-## When a disk is fuller than this factor, we will not allow more data to be
-## written to the system, unless this data is written in order to reduce
-## storage consumption, such as resizing files to become smaller or add
-## meta entries to write remove entries into. This value is set high as
-## default as we expect a lot of users to have formatted their disks to
-## already reserve 8% of the data to root user which is often default. We
-## suggest using 0% reserved for root, and rather set this parameter lower
-## to reserve space. That way, VDS have more space available in worst case
-## in order to resize files to become smaller.
-## DEPRECATED: see stor-memfilepersistence config instead
-disk_full_factor double default=0.98 restart
-
-## The grow factor sets how much free space to add to a file when we resize 
-## it, whether we are making it smaller or larger. If the space we need
-## after the operation triggering the resize is X, then the file will be
-## resized to be of size X * growfactor. In a growing system with no
-## deletes, a growfactor of 2 will make the files have 25% free space on
-## average. Reducing it to 1.5 will reduce the average free space to 16.7%.
-## DEPRECATED: see stor-memfilepersistence config instead
-grow_factor double default=2.0
-
-## If files fall below this fill rate, resize file to become smaller.
-## Note that this parameter is tightly coupled with the growfactor. For
-## instance, with a growfactor of 2.0, a file will contain 50 % free space
-## after a resize. If the min fill rate then is 0.50, that means that if a
-## single doc is deleted from this file, we need to resize it to
-## become smaller straight away.
-## DEPRECATED: see stor-memfilepersistence config instead
-min_fill_rate double default=0.1
-
-## Minimum part of defragmented space one need to reclaim to allow
-## defragmentation of file. This value is given as a ratio of reclaimed
-## space compared to the total size of the data block.
-## Example: A body block of 100 MB, has 15 MB free space, with largest
-## continuos free space of 5 MB. Gain of defragmentation will then be 0.1.
-## DEPRECATED: see stor-memfilepersistence config instead
-defrag_minimum_gain double default=1.0 restart
-
-## When creating/resizing slotfiles, one uses average document sizes to
-## decide how much free space to add to metadata, header and body portions.
-## This option can be used to allocate extra free space to meta data in
-## order to reduce the chance of the file needing resize due to lack of free
-## meta data entries.
-## DEPRECATED: see stor-memfilepersistence config instead
-overrepresent_meta_data_factor double default=1.2
-
-## When creating/resizing slotfiles, one uses average document sizes to
-## decide how much free space to add to metadata, header and body portions.
-## This option can be used to allocate extra free space to header data in
-## order to reduce the chance of the file needing resize due to lack of
-## header block space.
-## DEPRECATED: see stor-memfilepersistence config instead
-overrepresent_header_block_factor double default=1.1
-
-## Load types to use cache for. If empty, cache for all.
-## DEPRECATED: see stor-memfilepersistence config instead
-load_types_to_cache[] string  restart