Re: [PATCH v2 1/3] remote-curl: add testing for intelligent retry for HTTP

From: Junio C Hamano <gitster@pobox.com>
Date: Oct 13, 2020, 20:40 UTC
Message-ID: <xmqqblh5kggd.fsf@gitster.c.googlers.com>
In-Reply-To: <20201013191729.2524700-1-smcallis@google.com>

Sean McAllister <smcallis@google.com> writes:

Show 6 quoted lines

> +# generate a process unique one-up value
> +global_counter_for_nonce=0
> +gen_nonce () {
> +	global_counter_for_nonce=$((global_counter_for_nonce + 1)) &&
> +	echo "$global_counter_for_nonce"
> +}

This must not be called in a subprocess if the caller truly wants uniqueness. May want to be described in a comment.

Show 25 quoted lines

> diff --git a/t/lib-httpd/error-ntime.sh b/t/lib-httpd/error-ntime.sh
> new file mode 100755
> index 0000000000..64dc878746
> --- /dev/null
> +++ b/t/lib-httpd/error-ntime.sh
> @@ -0,0 +1,80 @@
> +#!/bin/sh
> +
> +# Script to simulate a transient error code with Retry-After header set.
> +#
> +# PATH_INFO must be of the form /<nonce>/<times>/<retcode>/<retry-after>/<path>
> +#   (eg: /dc724af1/3/429/10/some/url)
> +#
> +# The <nonce> value uniquely identifies the URL, since we're simulating
> +# a stateful operation using a stateless protocol, we need a way to "namespace"
> +# URLs so that they don't step on each other.
> +#
> +# The first <times> times this endpoint is called, it will return the given
> +# <retcode>, and if the <retry-after> is non-negative, it will set the
> +# Retry-After head to that value.
> +#
> +# Subsequent calls will return a 302 redirect to <path>.
> +#
> +# Supported error codes are 429,502,503, and 504
> +

I thought "error codes" were rephrased after the first round's review to some other term (which I do not recall--was it status?)?

Show 16 quoted lines

> +print_status() {
> +	if [ "$1" -eq "302" ]; then
> +		printf "Status: 302 Found\n"
> +	elif [ "$1" -eq "429" ]; then
> +		printf "Status: 429 Too Many Requests\n"
> +	elif [ "$1" -eq "502" ]; then
> +		printf "Status: 502 Bad Gateway\n"
> +	elif [ "$1" -eq "503" ]; then
> +		printf "Status: 503 Service Unavailable\n"
> +	elif [ "$1" -eq "504" ]; then
> +		printf "Status: 504 Gateway Timeout\n"
> +	else
> +		printf "Status: 500 Internal Server Error\n"
> +	fi
> +	printf "Content-Type: text/plain\n"
> +}

Style (Documentation/CodingGuidelines).

	print_status () {
		if test "$1" = "302"
		then
			printf "...";
		...
	}

but in this particular case, I do not see why we want if/else cascade. Perhaps

	print_status () {
		case "$1" in
		302)	printf "Status: 302 Found\n" ;;
		429)	... ;;
		...
		*)	printf "Status: 500 Internal Server Error\n" ;;
		esac
		printf "Content-Type: text/plain\n";
	}

would be more standard?

Also I am not sure why we want "printf ...\n" not "echo" here. If we want to talk HTTP ourselves strictly, I would understand avoiding "echo" and doing "printf ...\r\n", though. If we fear DOS line endings coming out of localized "echo", and ensure we use LF line ending even on Windows and Cygwin, it is sort of understandable but if that is what is going on, that does not explain a lone "echo" at the end of the caller.

Puzzled.

  +oIFS="$IFS"

> +IFS='/'
> +set -f
> +set -- $PATH_INFO
> +set +f

  +IFS="$oIFS"

Show 15 quoted lines

> +
> +# pull out first four path components
> +shift
> +nonce=$1
> +times=$2
> +code=$3
> +retry=$4
> +shift 4
> +
> +# glue the rest back together as redirect path
> +path=""
> +while [ "$#" -gt "0" ]; do
> +	path="${path}/$1"
> +	shift
> +done

Hmph.  Would this work better, I wonder?

	path=${PATH_INFO#*/}	;# discard leading '/'
	nonce=${path%%/*}	path=${path#*/}
	times=${path%%/*}	path=${path#*/}
	code=${path%%/*}	path=${path#*/}
	retry=${path%%/*}	path=${path#*/}

At least it is shorter and easier to read.

Show 6 quoted lines

> +# leave a cookie for this request/retry count
> +state_file="request_${REMOTE_ADDR}_${nonce}_${times}_${code}_${retry}"
> +
> +if [ ! -f "$state_file" ]; then
> +	echo 0 > "$state_file"
> +fi

Style (Documentation/CodingGuidelines, look for "For shell scripts specifically").

 - use "test" not "[]";

 - don't write ";then" on the same line (rule of thumb. you should
   be able to write your shell scripts without semicolon except for
   double-semicolons in case/esac statements)

 - don't leave SP between redirection operator '>' and its target
   file, i.e. write 'echo 0 >"$state_file"'.

Show 16 quoted lines

> +read -r cnt < "$state_file"
> +if [ "$cnt" -lt "$times" ]; then
> +	echo $((cnt+1)) > "$state_file"
> +
> +	# return error
> +	print_status "$code"
> +	if [ "$retry" -ge "0" ]; then
> +		printf "Retry-After: %s\n" "$retry"
> +	fi
> +else
> +	# redirect
> +	print_status 302
> +	printf "Location: %s?%s\n" "$path" "${QUERY_STRING}"
> +fi
> +
> +echo

This "echo" to the client also puzzles me further, after seeing puzzling use of "printf ...\n" earlier.

Show 13 quoted lines

> diff --git a/t/t5601-clone.sh b/t/t5601-clone.sh
> index 7df3c5373a..72aaed5a93 100755
> --- a/t/t5601-clone.sh
> +++ b/t/t5601-clone.sh
> @@ -756,6 +756,15 @@ test_expect_success 'partial clone using HTTP' '
>  	partial_clone "$HTTPD_DOCUMENT_ROOT_PATH/server" "$HTTPD_URL/smart/server"
>  '
>  
> +test_expect_success 'partial clone using HTTP with redirect' '
> +	_NONCE=`gen_nonce` && export _NONCE &&
> +	curl "$HTTPD_URL/error_ntime/${_NONCE}/3/502/10/smart/server" &&
> +	curl "$HTTPD_URL/error_ntime/${_NONCE}/3/502/10/smart/server" &&
> +	curl "$HTTPD_URL/error_ntime/${_NONCE}/3/502/10/smart/server" &&

Why do we need to test "curl" here? Is this remnant of debugging of the server side?

Show 6 quoted lines

> +	partial_clone "$HTTPD_DOCUMENT_ROOT_PATH/server" "$HTTPD_URL/error_ntime/${_NONCE}/3/502/10/smart/server"
> +'
> +
> +
>  # DO NOT add non-httpd-specific tests here, because the last part of this
>  # test script is only executed when httpd is available and enabled.