lib/google/cloud/bigtable/v2/doc/google/bigtable/v2/data.rb in google-cloud-bigtable-0.1.2 vs lib/google/cloud/bigtable/v2/doc/google/bigtable/v2/data.rb in google-cloud-bigtable-0.1.3
- old
+ new
@@ -34,11 +34,11 @@
# @!attribute [rw] name
# @return [String]
# The unique key which identifies this family within its row. This is the
# same key that's used to identify the family in, for example, a RowFilter
# which sets its "family_name_regex_filter" field.
- # Must match +[-_.a-zA-Z0-9]++, except that AggregatingRowProcessors may
+ # Must match `[-_.a-zA-Z0-9]+`, except that AggregatingRowProcessors may
# produce cells in a sentinel family with an empty name.
# Must be no greater than 64 characters in length.
# @!attribute [rw] columns
# @return [Array<Google::Bigtable::V2::Column>]
# Must not be empty. Sorted in order of increasing "qualifier".
@@ -48,11 +48,11 @@
# table.
# @!attribute [rw] qualifier
# @return [String]
# The unique key which identifies this column within its family. This is the
# same key that's used to identify the column in, for example, a RowFilter
- # which sets its +column_qualifier_regex_filter+ field.
+ # which sets its `column_qualifier_regex_filter` field.
# May contain any byte string, including the empty string, up to 16kiB in
# length.
# @!attribute [rw] cells
# @return [Array<Google::Bigtable::V2::Cell>]
# Must not be empty. Sorted in order of decreasing "timestamp_micros".
@@ -64,11 +64,11 @@
# The cell's stored timestamp, which also uniquely identifies it within
# its column.
# Values are always expressed in microseconds, but individual tables may set
# a coarser granularity to further restrict the allowed values. For
# example, a table which specifies millisecond granularity will only allow
- # values of +timestamp_micros+ which are multiples of 1000.
+ # values of `timestamp_micros` which are multiples of 1000.
# @!attribute [rw] value
# @return [String]
# The value stored in the cell.
# May contain any byte string, including the empty string, up to 100MiB in
# length.
@@ -157,22 +157,22 @@
# There are two broad categories of RowFilters (true filters and transformers),
# as well as two ways to compose simple filters into more complex ones
# (chains and interleaves). They work as follows:
#
# * True filters alter the input row by excluding some of its cells wholesale
- # from the output row. An example of a true filter is the +value_regex_filter+,
+ # from the output row. An example of a true filter is the `value_regex_filter`,
# which excludes cells whose values don't match the specified pattern. All
# regex true filters use RE2 syntax (https://github.com/google/re2/wiki/Syntax)
# in raw byte mode (RE2::Latin1), and are evaluated as full matches. An
- # important point to keep in mind is that +RE2(.)+ is equivalent by default to
- # +RE2([^\n])+, meaning that it does not match newlines. When attempting to
- # match an arbitrary byte, you should therefore use the escape sequence +\C+,
- # which may need to be further escaped as +\\C+ in your client language.
+ # important point to keep in mind is that `RE2(.)` is equivalent by default to
+ # `RE2([^\n])`, meaning that it does not match newlines. When attempting to
+ # match an arbitrary byte, you should therefore use the escape sequence `\C`,
+ # which may need to be further escaped as `\\C` in your client language.
#
# * Transformers alter the input row by changing the values of some of its
# cells in the output, without excluding them completely. Currently, the only
- # supported transformer is the +strip_value_transformer+, which replaces every
+ # supported transformer is the `strip_value_transformer`, which replaces every
# cell's value with the empty string.
#
# * Chains and interleaves are described in more detail in the
# RowFilter.Chain and RowFilter.Interleave documentation.
#
@@ -248,61 +248,61 @@
# In this case we have a duplicate with column "A:B" and timestamp 2,
# because one copy passed through the all filter while the other was
# passed through the label and sink. Note that one copy has label "foo",
# while the other does not.
#
- # Cannot be used within the +predicate_filter+, +true_filter+, or
- # +false_filter+ of a {Google::Bigtable::V2::RowFilter::Condition Condition}.
+ # Cannot be used within the `predicate_filter`, `true_filter`, or
+ # `false_filter` of a {Google::Bigtable::V2::RowFilter::Condition Condition}.
# @!attribute [rw] pass_all_filter
# @return [true, false]
# Matches all cells, regardless of input. Functionally equivalent to
- # leaving +filter+ unset, but included for completeness.
+ # leaving `filter` unset, but included for completeness.
# @!attribute [rw] block_all_filter
# @return [true, false]
# Does not match any cells, regardless of input. Useful for temporarily
# disabling just part of a filter.
# @!attribute [rw] row_key_regex_filter
# @return [String]
# Matches only cells from rows whose keys satisfy the given RE2 regex. In
# other words, passes through the entire row when the key matches, and
# otherwise produces an empty row.
- # Note that, since row keys can contain arbitrary bytes, the +\C+ escape
- # sequence must be used if a true wildcard is desired. The +.+ character
- # will not match the new line character +\n+, which may be present in a
+ # Note that, since row keys can contain arbitrary bytes, the `\C` escape
+ # sequence must be used if a true wildcard is desired. The `.` character
+ # will not match the new line character `\n`, which may be present in a
# binary key.
# @!attribute [rw] row_sample_filter
# @return [Float]
# Matches all cells from a row with probability p, and matches no cells
# from the row with probability 1-p.
# @!attribute [rw] family_name_regex_filter
# @return [String]
# Matches only cells from columns whose families satisfy the given RE2
- # regex. For technical reasons, the regex must not contain the +:+
+ # regex. For technical reasons, the regex must not contain the `:`
# character, even if it is not being used as a literal.
# Note that, since column families cannot contain the new line character
- # +\n+, it is sufficient to use +.+ as a full wildcard when matching
+ # `\n`, it is sufficient to use `.` as a full wildcard when matching
# column family names.
# @!attribute [rw] column_qualifier_regex_filter
# @return [String]
# Matches only cells from columns whose qualifiers satisfy the given RE2
# regex.
- # Note that, since column qualifiers can contain arbitrary bytes, the +\C+
- # escape sequence must be used if a true wildcard is desired. The +.+
- # character will not match the new line character +\n+, which may be
+ # Note that, since column qualifiers can contain arbitrary bytes, the `\C`
+ # escape sequence must be used if a true wildcard is desired. The `.`
+ # character will not match the new line character `\n`, which may be
# present in a binary qualifier.
# @!attribute [rw] column_range_filter
# @return [Google::Bigtable::V2::ColumnRange]
# Matches only cells from columns within the given range.
# @!attribute [rw] timestamp_range_filter
# @return [Google::Bigtable::V2::TimestampRange]
# Matches only cells with timestamps within the given range.
# @!attribute [rw] value_regex_filter
# @return [String]
# Matches only cells with values that satisfy the given regular expression.
- # Note that, since cell values can contain arbitrary bytes, the +\C+ escape
- # sequence must be used if a true wildcard is desired. The +.+ character
- # will not match the new line character +\n+, which may be present in a
+ # Note that, since cell values can contain arbitrary bytes, the `\C` escape
+ # sequence must be used if a true wildcard is desired. The `.` character
+ # will not match the new line character `\n`, which may be present in a
# binary value.
# @!attribute [rw] value_range_filter
# @return [Google::Bigtable::V2::ValueRange]
# Matches only cells with values that fall within the given range.
# @!attribute [rw] cells_per_row_offset_filter
@@ -316,13 +316,13 @@
# If duplicate cells are present, as is possible when using an Interleave,
# each copy of the cell is counted separately.
# @!attribute [rw] cells_per_column_limit_filter
# @return [Integer]
# Matches only the most recent N cells within each column. For example,
- # if N=2, this filter would match column +foo:bar+ at timestamps 10 and 9,
- # skip all earlier cells in +foo:bar+, and then begin matching again in
- # column +foo:bar2+.
+ # if N=2, this filter would match column `foo:bar` at timestamps 10 and 9,
+ # skip all earlier cells in `foo:bar`, and then begin matching again in
+ # column `foo:bar2`.
# If duplicate cells are present, as is possible when using an Interleave,
# each copy of the cell is counted separately.
# @!attribute [rw] strip_value_transformer
# @return [true, false]
# Replaces each cell's value with the empty string.
@@ -331,16 +331,16 @@
# Applies the given label to all cells in the output row. This allows
# the client to determine which results were produced from which part of
# the filter.
#
# Values must be at most 15 characters in length, and match the RE2
- # pattern +[a-z0-9\\-]++
+ # pattern `[a-z0-9\\-]+`
#
# Due to a technical limitation, it is not currently possible to apply
# multiple labels to a cell. As a result, a Chain may have no more than
- # one sub-filter which contains a +apply_label_transformer+. It is okay for
- # an Interleave to contain multiple +apply_label_transformers+, as they
+ # one sub-filter which contains a `apply_label_transformer`. It is okay for
+ # an Interleave to contain multiple `apply_label_transformers`, as they
# will be applied to separate copies of the input. This may be relaxed in
# the future.
class RowFilter
# A RowFilter which sends rows through several RowFilters in sequence.
# @!attribute [rw] filters
@@ -388,19 +388,19 @@
# true and false filters, which may lead to inconsistent or unexpected
# results. Additionally, Condition filters have poor performance, especially
# when filters are set for the false condition.
# @!attribute [rw] predicate_filter
# @return [Google::Bigtable::V2::RowFilter]
- # If +predicate_filter+ outputs any cells, then +true_filter+ will be
- # evaluated on the input row. Otherwise, +false_filter+ will be evaluated.
+ # If `predicate_filter` outputs any cells, then `true_filter` will be
+ # evaluated on the input row. Otherwise, `false_filter` will be evaluated.
# @!attribute [rw] true_filter
# @return [Google::Bigtable::V2::RowFilter]
- # The filter to apply to the input row if +predicate_filter+ returns any
+ # The filter to apply to the input row if `predicate_filter` returns any
# results. If not provided, no results will be returned in the true case.
# @!attribute [rw] false_filter
# @return [Google::Bigtable::V2::RowFilter]
- # The filter to apply to the input row if +predicate_filter+ does not
+ # The filter to apply to the input row if `predicate_filter` does not
# return any results. If not provided, no results will be returned in the
# false case.
class Condition; end
end
@@ -420,11 +420,11 @@
class Mutation
# A Mutation which sets the value of the specified cell.
# @!attribute [rw] family_name
# @return [String]
# The name of the family into which new data should be written.
- # Must match +[-_.a-zA-Z0-9]++
+ # Must match `[-_.a-zA-Z0-9]+`
# @!attribute [rw] column_qualifier
# @return [String]
# The qualifier of the column into which new data should be written.
# Can be any byte string, including the empty string.
# @!attribute [rw] timestamp_micros
@@ -442,11 +442,11 @@
# A Mutation which deletes cells from the specified column, optionally
# restricting the deletions to a given timestamp range.
# @!attribute [rw] family_name
# @return [String]
# The name of the family from which cells should be deleted.
- # Must match +[-_.a-zA-Z0-9]++
+ # Must match `[-_.a-zA-Z0-9]+`
# @!attribute [rw] column_qualifier
# @return [String]
# The qualifier of the column from which cells should be deleted.
# Can be any byte string, including the empty string.
# @!attribute [rw] time_range
@@ -456,11 +456,11 @@
# A Mutation which deletes all cells from the specified column family.
# @!attribute [rw] family_name
# @return [String]
# The name of the family from which cells should be deleted.
- # Must match +[-_.a-zA-Z0-9]++
+ # Must match `[-_.a-zA-Z0-9]+`
class DeleteFromFamily; end
# A Mutation which deletes all cells from the containing row.
class DeleteFromRow; end
end
@@ -468,23 +468,23 @@
# Specifies an atomic read/modify/write operation on the latest value of the
# specified column.
# @!attribute [rw] family_name
# @return [String]
# The name of the family to which the read/modify/write should be applied.
- # Must match +[-_.a-zA-Z0-9]++
+ # Must match `[-_.a-zA-Z0-9]+`
# @!attribute [rw] column_qualifier
# @return [String]
# The qualifier of the column to which the read/modify/write should be
# applied.
# Can be any byte string, including the empty string.
# @!attribute [rw] append_value
# @return [String]
- # Rule specifying that +append_value+ be appended to the existing value.
+ # Rule specifying that `append_value` be appended to the existing value.
# If the targeted cell is unset, it will be treated as containing the
# empty string.
# @!attribute [rw] increment_amount
# @return [Integer]
- # Rule specifying that +increment_amount+ be added to the existing value.
+ # Rule specifying that `increment_amount` be added to the existing value.
# If the targeted cell is unset, it will be treated as containing a zero.
# Otherwise, the targeted cell must contain an 8-byte value (interpreted
# as a 64-bit big-endian signed integer), or the entire request will fail.
class ReadModifyWriteRule; end
end
\ No newline at end of file