Re: [PHP-DEV] Need help with strange and random bug

[Andrey E Baranov]

Nikita Popov at 2017-02-05 21:44 wrote:
On Sun, Feb 5, 2017 at 3:03 PM, Andrey E Baranov <and...@andb.name> 
wrote:

Hi all,

I have problem, debug is not yet finished, and for now do not can
reproduce, problem occurs only in production (after migrate to PHP7).
perhaps someone else can help :)
PHP 7.0.13 (Ubuntu 16.04 package repository), php-fpm

For simplify:
```
...
Log::debug('1:' . print_r($video, true));
Log::debug('2:' . json_encode($video) . ' json_last_error: ' .
json_last_error() . ' json_last_error_msg: ' . json_last_error_msg());

$video['delivery'] = [2]; // after this line: serialize and 
json_encode do
not work as expected

Log::debug('3:' . print_r($video, true));
Log::debug('4:' . serialize($video));
Log::debug('5:' . json_encode($video) . ' json_last_error: ' .
json_last_error() . ' json_last_error_msg: ' . json_last_error_msg());
...
```

log with bug:
```
DEBUG: 1:Array
(
    [mimes] => Array
        (
            [0] => video/mp4
        )

    [linearity] => 1
    [minduration] => 5
    [maxduration] => 60
    [protocols] => Array
        (
            [0] => 2
            [1] => 3
        )

    [w] => 818
    [h] => 460
    [minbitrate] => 100
    [maxbitrate] => 1200
)
DEBUG: 2: {"mimes":["video/mp4"],"linearity":1,"minduration":5,"maxdur
ation":60,"protocols":[2,3],"w":818,"h":460,"minbitrate":
100,"maxbitrate":1200}
DEBUG: 3:Array
(
    [mimes] => Array
        (
            [0] => video/mp4
        )

    [linearity] => 1
    [minduration] => 5
    [maxduration] => 60
    [protocols] => Array
        (
            [0] => 2
            [1] => 3
        )

    [w] => 818
    [h] => 460
    [minbitrate] => 100
    [maxbitrate] => 1200
    [delivery] => Array
        (
            [0] => 2
        )

)
DEBUG: 4:a:10:{s:5:"mimes";a:1:{i:0;s:9:"video/mp4";}s:9:"linearity
";i:1;s:11:"minduration";i:5;s:11:"maxduration";i:60;s:9:"p
rotocols";a:2:{i:0;i:2;i:1;i:3;}s:1:"w";i:818;s:1:"h";i:460;
s:10:"minbitrate";i:100;s:10:"maxbitrate";i:1200;s:8:"delivery";N;}
DEBUG: 5: json_last_error: 6 json_last_error_msg: Recursion detected
```

- 4: `serialize($video)` - `"delivery";N;` - it is mean that 
`delivery` is
Null - that is not true
- 5: `json_encode($video)` - Recursion detected

Problem occurs in random period after start fpm process (after 10..30
min), and remains constant until restart fpm

Same problem as I see in http://stackoverflow.com/quest
ions/37456845/what-does-json-error-recursion-mean-from-json-encode

After deep analyze:

php_json_encode_array:
```
```
if (myht && ZEND_HASH_GET_APPLY_COUNT(myht) > 1) {
  JSON_G(error_code) = PHP_JSON_ERROR_RECURSION;
  smart_str_appendl(buf, "null", 4);
  return;
}
```

serialize:
```
if (... || (Z_TYPE_P(data) == IS_ARRAY && 
Z_ARRVAL_P(data)->u.v.nApplyCount
> 1)
  ) {
    smart_str_appendl(buf, "N;", 2);
...
```

so looks like problem in `HashTable.v.nApplyCount`


Also I am find commit - https://github.com/php/php-src
/commit/d26ca894020bc28eb55b153fac5548374d5ce16d

In this commit added
```
target->u.flags = (source->u.flags & 
~(...|ZEND_HASH_APPLY_COUNT_MASK))...
```
But not added to case if `if (GC_FLAGS(source) & IS_ARRAY_IMMUTABLE)` 
-
maybe need fix in this place? :) I guess that the problem in immutable
arrays.


The apply count on immutable arrays is always zero, so this shouldn't 
be a
problem. However, it may be that some other code is incrementing the 
apply
count on an immutable array. Can you please check whether running with
opcache.protect_memory=1 gives you any segmentation faults or bus 
errors?

Nikita

Nikita thanks! looks like `opcache.protect_memory=1` is usefull :)

```
#0  phongo_bson_append (bson=bson@entry=0x7fff0d47e220,
    flags=flags@entry=PHONGO_BSON_NONE, key=0x7fabe79ec070 "$in", 
key_len=3,
    entry=entry@entry=0x7fabe7f79a00) at 
/tmp/pear/temp/mongodb/src/bson.c:1027
...
```

and it is fixed in
https://github.com/mongodb/mongo-php-driver/commit/129753881a223b8378fc7b8137104c65d8d8828f

so, with PHP7 need use mongodb >= 1.1.10 (as minimum :) we missed that 
mongodb is outdated in our production, but the relationship between our 
bug and mongodb was not the obvious)

--
PHP Internals - PHP Runtime Development Mailing List
To unsubscribe, visit: http://www.php.net/unsub.php